Gemini Models Explained: Flash vs Pro — and What Google AI Ultra Really Is

Human Verified 0

Share to: , and

A clean, wide illustration of a branching model family tree — one glowing central core splitting into three distinct coloured streams (deep blue, bright teal, warm amber) over a dark, minimal background. No text, no Google logo.

Gemini Models Explained: Flash vs Pro — and What Google AI Ultra Really Is

If you have tried to choose a Gemini model recently, you have probably run into a confusing lineup. The names sound similar — Flash, Pro, Flash-Lite, and now Carbon, Argon and Hydrogen — the prices change on a schedule, and half the roundups on the internet still describe a lineup that no longer exists. Worse, the most common comparison on the web, “Flash vs Pro vs Ultra”, treats three unlike things as three models. There is no Ultra model. Google AI Ultra is a subscription plan that unlocks models; it is not a model itself. This guide corrects that premise first, then compares the models that actually exist, with the prices, context windows and limits Google publishes on its own pages. Prices and model details checked 13 November 2026 against Google’s official model catalogue, pricing page and plans page. Prices are stated as of that check, and two price rises are already scheduled — Gemini 4 Carbon and Argon double on 1 December 2026, and the two current Flash models double on 1 January 2027 — so the dates matter as much as the numbers.

The short version, before the detail, is this. Gemini 3.8 Flash is the default choice for most work: the most intelligent Flash model Google has shipped, stable, with a free tier, and cheap enough that most sessions cost a few cents. Gemini 3.1 Pro is the flagship for the hardest work — complex software engineering, agentic workflows that must not drift, and multimodal reasoning — and it costs roughly three times as much per output token. Flash-Lite is the cheapest way to run Gemini at volume, for translation, extraction and subagent work. Gemini 4 is the newest generation: Hydrogen for everyday coding, Argon for efficient frontier reasoning, Carbon for the heaviest reasoning. And Google AI Ultra, at $99.99 a month, is the plan that buys the highest access to those newest models, plus a large bundle of storage, credits and extras. Pick the model for the task, and pick the plan for how much of it you need.

A clean, wide illustration of a branching model family tree — one glowing central core splitting into three distinct coloured streams (deep blue, bright teal, warm amber) over a dark, minimal background. No text, no Google logo.

The Short Answer: Which Gemini Model Should You Use?

<img src="

A person at a desk choosing between three clearly different doors/passages, warm daylight, minimal composition.

” alt=”A person at a desk choosing between three clearly different doors/passages, warm daylight, minimal composition.”>

Start with the decision, because the rest of this guide is the evidence for it. If you are doing everyday work — chatting, drafting, summarising, light coding — use Gemini 4 Hydrogen or Gemini 3.8 Flash, and if you are on the free tier of the Gemini app or paying nothing through the API, Flash-Lite will do most of it for almost nothing. If you are building something where a mistake is expensive — a long software-engineering task, an agent that has to use tools reliably across many steps, or a multimodal task that mixes documents, images and code — use Gemini 3.1 Pro, or Gemini 4 Carbon if you have access to it through an Ultra plan. If you are processing volume — thousands of documents, a translation pipeline, a fleet of subagents — use Flash-Lite and spend the savings on the calls that actually need a bigger model. That is the whole picker in one paragraph. Everything below explains why.

The most important fact about the lineup is how little separates the models on raw capacity. Every current text model in the Gemini family has a 1-million-token input context window and a 65,536-token maximum output, with one exception: the Gemini 4 models can generate up to 1 million tokens in a single long decode. Context window, the number most people ask about first, does not differentiate these models at all. What differentiates them is reasoning depth, speed, price and availability — and price is where the differences are starkest. Gemini 3.1 Pro costs $2.00 per million input tokens and $12.00 per million output tokens on prompts up to 200,000 tokens; Gemini 3.1 Flash-Lite costs $0.25 and $1.50. The flagship is eight times the price of the cheapest model, for the same context window. You are paying for what the model does with that context, not for the size of it.

A second fact shapes every recommendation in this guide: the lineup moved fast in 2026, and the names in most older articles are out of date. The current Pro is Gemini 3.1 Pro, not the original Gemini 3 Pro, which no longer appears in Google’s model catalogue at all. The current Flash is Gemini 3.8 Flash, a stable model updated in September 2026; Gemini 3.6 Flash is the previous generation and the original Gemini 3 Flash is a legacy preview model. Gemini 4 — Carbon, Argon and Hydrogen — is a new generation that sits alongside the 3.x family, not a replacement for it. And the two Gemini 2.5 models that many tutorials still name are deprecated, with automatic re-routing to newer models from 1 December 2026. If a tutorial, a code sample or a roundup names Gemini 2.5 Pro or the original Gemini 3 Pro as the current choice, it is describing a lineup that has moved on. This guide is written against the lineup as Google publishes it on 13 November 2026, and the final section explains how to keep it current.

One more correction belongs up front, because it is the premise of this whole article. “Ultra” is not a model, and there is no three-way comparison between Flash, Pro and Ultra. Google AI Ultra is a $99.99-a-month subscription plan. It unlocks the highest access to Gemini 4 Carbon and Gemini 4 Argon, higher access to Gemini 3.1 Pro, and a bundle of storage, credits and services. A plan that unlocks models cannot itself be a model, and comparing it against Flash and Pro as though it were one would be comparing a ticket with the trains it lets you ride. Section 11 of this guide explains exactly what Ultra is, what it costs, and who should and should not pay for it. If you arrived here looking for the “Ultra model”, that section is the answer — and for readers who are new to AI chatbots generally, our guide to how to use ChatGPT is a useful companion for the basics of working with an AI assistant, whatever model sits behind it.

The Gemini Family in One Table

An orderly row of labelled-free glass containers of different sizes on a workbench, suggesting a comparison at a glance.

The table below is the whole lineup as Google publishes it, in one place. Prices are the Standard tier of the Gemini API, in US dollars per million tokens, as of 13 November 2026. Read the footnotes under the table before you quote any number from it, because four of these prices are introductory and change within weeks of this guide’s publication.

ModelAPI codeStatusInput context / Max outputInput $ / 1M tokensOutput $ / 1M tokensFree tier
Gemini 4 Carbongemini-4-carbonCurrent — Gemini 4 flagship reasoning1M / 65K–1M*$2.00†$10.00†No
Gemini 4 Argongemini-4-argonCurrent — efficient frontier reasoning1M / 65K–1M*$2.00†$10.00†No
Gemini 4 Hydrogengemini-4-hydrogenCurrent — everyday coding and agentic work1M / 65K–1M*$1.50$7.50Yes
Gemini 3.1 Progemini-3.1-pro-previewPreview — the current Pro1,048,576 / 65,536$2.00 / $4.00‡$12.00 / $18.00‡No§
Gemini 3.8 Flashgemini-3.8-flashStable — most intelligent Flash1,048,576 / 65,536$0.75¶$3.75¶Yes
Gemini 3.6 Flashgemini-3.6-flashStable — previous-generation Flash1,048,576 / 65,536$0.75¶$3.75¶Yes
Gemini 3 Flash (original)gemini-3-flash-previewPreview — legacy FlashNot confirmed in this guide‖$0.50 ($1.00 audio)$3.00Yes
Gemini 3.5 Flash-Litegemini-3.5-flash-liteStable1,048,576 / 65,536$0.30$2.50Yes
Gemini 3.1 Flash-Litegemini-3.1-flash-liteStable — cheapest current text model1,048,576 / 65,536$0.25 ($0.50 audio)$1.50Yes

* Gemini 4 model cards list “Max output 65K / 1M tokens” — long-decode generation up to 1 million tokens. That is stated exactly as Google lists it, not simplified to a single number. † Gemini 4 Carbon and Argon introductory pricing: $2.00 input and $10.00 output per million tokens through 30 November 2026, then $4.00 and $20.00 from 1 December 2026. Both models show identical prices on Google’s pricing page, and this guide reports them as read — Argon is not listed as cheaper than Carbon. ‡ Gemini 3.1 Pro is priced by prompt size: prompts of up to 200,000 tokens cost $2.00 in and $12.00 out per million tokens; prompts over 200,000 tokens cost $4.00 in and $18.00 out. Context caching for 3.1 Pro costs $0.20 per million tokens for prompts up to 200,000 tokens and $0.40 above that, plus $4.50 per million tokens per hour of storage. § Gemini 3.1 Pro has no free API tier, but it can be tried free inside Google AI Studio. ¶ Gemini 3.8 Flash and Gemini 3.6 Flash introductory pricing: $0.75 in and $3.75 out per million tokens through 31 December 2026, then $1.50 and $7.50 from 1 January 2027. The two models carry identical prices. ‖ Google’s pricing page confirms the original Gemini 3 Flash price, but its model page was not among the pages read for this guide, so its exact token limits are not stated here rather than guessed.

Three patterns in that table are worth noticing before moving on. First, the 1-million-token input context and 65,536-token maximum output are the same across every model except for the Gemini 4 long-decode exception — uniformity, not variety, is the story of this generation. Second, the price spread is enormous for models with identical context windows: from $0.25 per million input tokens for Flash-Lite to $2.00 for Pro and the Gemini 4 flagships, and from $1.50 to $12.00 on output. Third, the free tier draws a clean line: the Flash family and Hydrogen are free to try through the API, while Carbon, Argon and 3.1 Pro are paid-only — though 3.1 Pro can be tested free inside AI Studio, which is the sensible place to try it before committing to a paid plan. In plain language, then: Flash-Lite is the volume worker, Flash is the all-rounder, Pro is the specialist for the hardest jobs, and Gemini 4 is the new generation that currently sits on top of the range for subscribers who can reach it.

One more pricing fact applies to every row of the table. Thinking tokens — the tokens a model spends reasoning before it answers — are billed as output tokens on every model listed. Google states the output price as “including thinking tokens”, which means a model that thinks longer costs more even when its final answer is short. That matters most for the reasoning-heavy models, where the thinking is the point: a long chain of thought on a hard coding problem is real work, and it is billed at the output rate. It also means the cheapest way to control cost is often not to pick a cheaper model but to pick a model whose thinking level matches the task — and for Gemini 3.8 Flash, Google lets you choose thinking levels of low, medium or high, with the “minimal” level not supported and returning an error if requested.

Gemini 3.1 Pro: The Flagship, Explained

A close-up of an engineer working through a complex blueprint with layered transparent sheets, focused and precise.

Gemini 3.1 Pro is the current Pro — the flagship of the Gemini 3 generation — and it is the model to reach for when the work is hard enough that a cheaper model’s mistakes would cost more than the price difference. Google describes it as built for multimodal understanding, agentic capabilities and vibe-coding, with better thinking, improved token efficiency and a more grounded, factually consistent experience than its predecessors. The phrasing that matters most for buyers is the second half of Google’s own description: it is optimised for software engineering behaviour and usability, and for agentic workflows that require precise tool use and reliable multi-step execution. In plain language, Pro is the model you choose when the task has many steps, each step depends on the last, and a wrong turn early on ruins the whole result. Its latest update, according to Google’s model page, was February 2026, and its model card is published by Google DeepMind.

Pro’s pricing is the only tiered pricing in the lineup, and it is worth understanding before you use it, because a long prompt costs twice as much per token as a short one. For prompts of up to 200,000 tokens, the Standard tier costs $2.00 per million input tokens and $12.00 per million output tokens. For prompts over 200,000 tokens, those rates rise to $4.00 and $18.00. That means a conversation that grows past 200,000 tokens — a long codebase pasted in, a large set of documents, a multi-step agent run that accumulates context — doubles in unit price partway through. Context caching softens that for repeated work: caching costs $0.20 per million tokens for prompts up to 200,000 tokens and $0.40 above that, plus $4.50 per million tokens per hour of storage. If you are sending the same large context repeatedly, caching is the difference between a bill that grows linearly and one that grows much more slowly. The Batch API, which costs half the Standard rate, brings a 3.1 Pro session down to $1.00 in and $6.00 out per million tokens for prompts up to 200,000 tokens — a real saving for work that does not need an immediate answer.

There is no free tier for Gemini 3.1 Pro through the API. You can, however, try it free inside Google AI Studio, which is the right place to test it before you commit to paid usage. That distinction matters for two kinds of reader. A developer evaluating whether Pro is worth the premium over Flash should run the same hard task through both models in AI Studio and compare the results before spending anything. A casual user who only ever opens the Gemini app will meet Pro through a subscription — expanded access comes with Google AI Pro, and higher access with Google AI Ultra — rather than through the API at all. Either way, the question to ask is not “is Pro better?” but “is Pro better at the specific thing I am doing, by enough to justify roughly three times the output price of Flash?” For everyday chat, drafting and summarising, the honest answer is usually no. For a multi-file refactoring, an agent that has to call tools in the right order ten times in a row, or a task where the model must hold a large, messy set of sources in mind and stay consistent, the answer is often yes — and the cost of the model is small next to the cost of fixing a cheaper model’s mistakes.

One capability note is worth flagging because it surprises people who assume the flagship does everything the cheaper models do. Computer use — the ability for a model to operate a computer interface directly — is supported on Gemini 3.8 Flash, 3.6 Flash and 3.5 Flash-Lite, but it is not listed for Gemini 3.1 Pro. The flagship is not a superset of every Flash capability; the two lines have different strengths, and the capability tables in Google’s own documentation are the authority on which model does what. Similarly, none of the text models in this guide — Pro included — can generate audio, generate images, or run the Live API for real-time voice. Those jobs belong to specialist models: Gemini 3.8 Flash TTS and Flash-Lite TTS for speech, the Nano Banana family for images, and Gemini 3.8 Live for real-time voice. Do not expect Pro to draw you a picture or speak aloud; that is not what it is for.

Gemini 3.8 Flash: The Smartest Flash Yet

A sprinter mid-stride on a track at dawn, motion blur on the background, sharp on the runner — speed with direction.

If there is a single default answer in this guide, it is Gemini 3.8 Flash. Google calls it its most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents and complex enterprise workflows — all with the speed and cost efficiency of Flash. It is a stable model, not a preview, and its latest update was September 2026. That combination — stable, most capable in its line, and priced at the Flash rate — is why it is the right starting point for most developers and most production use. When you are not sure which model to use, start here, and move up to Pro only when you can point to a specific failure that Pro would fix.

The pricing is straightforward, but it carries a date you need to know. Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens — but that is introductory pricing, running through 31 December 2026. From 1 January 2027, both rates double, to $1.50 and $7.50. The same is true of Gemini 3.6 Flash, which carries identical prices on the same schedule. That doubling does not make Flash expensive — even at $1.50 and $7.50 it remains well below Pro’s $2.00 and $12.00 — but it does mean that any cost estimate you build on the current rates needs the January date attached to it. The worked examples in section 9 of this guide use the current rates and say so; if you are reading this after 1 January 2027, halve nothing and double the Flash numbers.

Flash’s capability profile is broad. Like every current text model, it accepts text, image, video, audio and PDF input, produces text output, and supports code execution, context caching, file search, function calling, grounding with Google Search and Maps, structured outputs, thinking and URL context. It also supports computer use, which the flagship Pro does not list. And its thinking is adjustable: you can set thinking levels of low, medium or high, matching the reasoning effort to the task — light thinking for a quick extraction, heavy thinking for a hard problem. The one level you cannot set is “minimal”, which is not supported and returns an error. That adjustability is part of what makes Flash the practical default: you can spend less on easy tasks and more on hard ones without switching models.

The free tier is the other half of Flash’s case. Gemini 3.8 Flash has a free tier through the API, which means you can build, test and run small-scale work without a billing account at all. Combined with its stability and its position as the most intelligent Flash, that makes it the model for prototypes, side projects, and any production workload where the task is well understood and the failure cost is low. The question Flash forces you to answer is not whether it is good enough in general — for most tasks, it is — but whether your specific task is one of the few where Pro’s deeper reasoning earns its premium. Section 8 takes that comparison head-on.

Gemini 3.6 Flash and the Original Gemini 3 Flash: What Changed

Two generations of the same tool side by side on a table — an older, worn version next to a newer, sleeker one.

Gemini 3.6 Flash is the previous generation of Flash, and Google’s own note about it says so plainly: it is “a previous-generation Flash model that remains available for production use. For our newest Flash model, see Gemini 3.8 Flash.” That sentence is the whole story. 3.6 Flash is not deprecated, not broken and not withdrawn; it is simply the model before the current one, kept available so that existing production systems do not have to migrate overnight. Google describes it as offering sustained frontier-level intelligence optimised for real-world tasks at a higher speed and lower cost, excelling at code generation, agentic execution and spatial reasoning, and as particularly effective for rapid agentic loops. Its latest update was July 2026. If your system already runs on 3.6 Flash and works, there is no emergency. But if you are choosing a Flash model today, there is no reason to pick the previous generation over the current one: 3.8 Flash costs exactly the same — $0.75 in and $3.75 out per million tokens through 31 December 2026, then $1.50 and $7.50 — and it is the more intelligent model.

The original Gemini 3 Flash, API code gemini-3-flash-preview, sits a step further back. Google’s pricing page calls it “our legacy Flash model, providing baseline speed and intelligence”, and that word — legacy — is the signal. It is still in preview status, it still has a free tier, and it costs $0.50 per million input tokens and $3.00 per million output tokens, with audio input at $1.00. It is the cheapest full Flash model in the lineup, cheaper than 3.8 Flash on both input and output. But cheap is not the same as good value. The original 3 Flash is the baseline against which the newer Flash models improved, and the improvements in 3.8 Flash — longer-horizon reasoning, stronger agentic performance, stable status rather than preview — are exactly the qualities that matter in production. The only honest reason to choose the original 3 Flash today is price sensitivity on a task so simple that the newer models’ advantages do not register. For anything else, the extra $0.25 per million input tokens for 3.8 Flash is money well spent.

There is a practical lesson in how Google handles these generations, and it applies to your own code as well as to your model choice. Google keeps older models available, labels them clearly as previous-generation or legacy, and lets you migrate on your own schedule — but the pricing page and the model catalogue always point at the newest model, and the newest model costs the same as the one before it. That is a deliberate pattern: when the successor costs no more than the predecessor, staying on the old model saves you nothing and costs you capability. The same pattern appears in the retirement section of this guide, where deprecated models are automatically re-routed to their successors. The rule of thumb is simple: use the newest stable model in the line unless you have a specific, tested reason not to. For Flash, that means 3.8.

Flash-Lite: The Cheapest Way to Run Gemini at Scale

A high-speed conveyor or sorting line moving many small identical packages, bright warehouse light.

Flash-Lite is the model most people overlook, and it is the one that changes the economics of a project more than any other. There are two current Flash-Lite models. Gemini 3.1 Flash-Lite costs $0.25 per million input tokens and $1.50 per million output tokens — the cheapest current text model Google offers, by a wide margin. Gemini 3.5 Flash-Lite costs $0.30 in and $2.50 out. Both are stable models with free tiers, both have the same 1,048,576-token input context and 65,536-token maximum output as the rest of the family, and both are built for the same job: high-volume, low-cost execution where latency and API cost are the primary constraints.

Google’s descriptions of the two models draw the distinction clearly. Gemini 3.1 Flash-Lite is a low-latency, cost-effective multimodal model optimised for high-frequency, lightweight tasks: high-volume agentic workflows, simple data extraction, and the places where the cost of each call matters more than the depth of any single answer. Its latest update was May 2026. Gemini 3.5 Flash-Lite is low-latency and cost-effective in the same way, but optimised for high-throughput, low-cost execution for subagent tasks and document parsing. Its latest update was July 2026. The difference between them is one of emphasis rather than kind: 3.1 Flash-Lite is the cheapest option in the family, and 3.5 Flash-Lite is the slightly newer, slightly dearer option tuned for parsing and subagent work. For most volume workloads either will do, and the price difference — five cents per million input tokens — is small enough that the choice should be driven by testing, not by the price tag.

The uses Google highlights for Flash-Lite are the best guide to when it belongs in your stack. Translation at scale, where the same operation repeats thousands of times and each instance is simple. Transcription, where the input is long and the output is a faithful rendering rather than a creative act. Lightweight agentic tasks and data extraction, where a small model pulls a structured answer out of a document. Document processing and summarisation, where the volume is the challenge and the reasoning is shallow. And one use that deserves more attention than it gets: model routing. Google’s own developer guide highlights that Gemini CLI uses Flash-Lite as the classifier that routes queries to Flash or Pro — a small, fast, cheap model deciding which bigger model should handle each question. That is the pattern to copy. Use Flash-Lite as the front door: let it classify, extract, translate and triage, and spend the Pro and Flash budget only on the questions that survive triage. A system built that way can be an order of magnitude cheaper than one that sends every question to the flagship.

Two capability details complete the picture. Gemini 3.5 Flash-Lite supports computer use, like the two current Flash models; Gemini 3.1 Flash-Lite does not list it. And both Flash-Lite models support the full shared capability set — multimodal input, code execution, context caching, file search, function calling, grounding, structured outputs, thinking and URL context — so the cheap model is not a crippled model. It is a model tuned for speed and cost on simple tasks, and the honest test of whether it fits your task is a direct one: run a sample of your real workload through Flash-Lite and through Flash, compare the outputs, and see whether the difference is worth three times the price on input and two and a half times on output. For translation, extraction and classification, the answer is very often no — and that no is worth real money at scale.

Gemini 4 Has Arrived: Carbon, Argon and Hydrogen in Plain Language

Three distinct mountain peaks at sunrise, one tallest, one mid-height, one broad and accessible — a new range on the horizon.

Gemini 4 is the newest generation of the Gemini family, and it arrives as three models with three distinct jobs. The names are elements — Carbon, Argon, Hydrogen — and the hierarchy is deliberate: Carbon is the strongest, Argon is the efficient frontier, Hydrogen is the everyday model for everyone. All three share the same 1-million-token input context and the same long-decode capability: their model cards list a maximum output of 65K to 1M tokens, meaning they can generate up to a million tokens in a single long decode, far beyond the 65,536-token ceiling of the 3.x models. That is the single biggest structural difference between the generations. Everything else — the pricing, the availability, the question of who can actually use them — follows from where each model sits in the hierarchy.

Gemini 4 Carbon is, in Google’s words, “our strongest reasoning model, built for complex software engineering, multimodal knowledge work, scientific reasoning, and cybersecurity defense.” Google describes it as state-of-the-art in cybersecurity and long-horizon capabilities. It is the model for the hardest reasoning work Google currently offers, and it is also the least accessible: there is no free tier, and in the consumer world it is available to Google AI Ultra subscribers in select countries. Its introductory price is $2.00 per million input tokens and $10.00 per million output tokens, running through 30 November 2026, after which both rates double to $4.00 and $20.00 from 1 December 2026. At the doubled rate, Carbon becomes the most expensive model in the lineup on output by a clear margin — which is the price of being the strongest reasoning model Google sells.

Gemini 4 Argon is “our efficient Gemini 4 frontier reasoning model, delivering faster responses and lower token spend per turn compared to Gemini 4 Carbon across complex coding, agentic, math, and science tasks.” Read that description carefully, because it contains a surprise: Argon is positioned as the efficient frontier model, faster and cheaper per turn than Carbon — yet Google’s pricing page lists identical prices for the two models, $2.00 in and $10.00 out through 30 November 2026, doubling to $4.00 and $20.00 from 1 December. This guide reports those prices exactly as Google lists them and does not “correct” them by assuming Argon must be cheaper. The efficiency Google describes is in token spend per turn — Argon reaches the answer with fewer tokens — rather than in the price per token. A model that uses fewer tokens to do the same job costs less in practice even at the same rate, and that is the sense in which Argon is the efficient choice. Argon has no free tier, and in the consumer world it is the frontier model that Google AI Pro subscribers get access to, with Ultra subscribers getting the highest access.

Gemini 4 Hydrogen is the accessible member of the family. Google describes it as bringing “cutting-edge intelligence to everyone, with speed and efficiency for everyday coding and agentic workflows, featuring a unified 1M-token context window and long-decode generation up to 1M tokens.” It costs $1.50 per million input tokens and $7.50 per million output tokens — notably cheaper than Carbon and Argon, and cheaper on output than 3.1 Pro’s $12.00 — and it is the only Gemini 4 model with a free tier. In the Gemini app, Hydrogen is the default-tier model for Google AI Plus subscribers. Hydrogen is the Gemini 4 model most readers will actually meet first, and for everyday coding and agentic work it is the natural entry point to the new generation: the long-decode capability, the 1M context, and a price below Pro, all without a subscription.

One point of framing matters for anyone deciding whether to move to Gemini 4 now. Gemini 4 is the newest family, but the 3.x family remains fully current. Gemini 3.1 Pro is still the current Pro, Gemini 3.8 Flash is still the stable default Flash, and the Flash-Lite models are still the cheapest way to run volume. Google has not retired the 3.x models to make room for the 4.x models; it has added a new tier on top. The practical consequence is that there is no pressure to migrate existing systems to Gemini 4, and the right question is narrower: does a specific Gemini 4 capability — the long decode, Carbon’s reasoning depth, Hydrogen’s price point — solve a problem your current model does not? If it does, move that workload. If it does not, the 3.x models you already use remain fully supported, and the price rises in December and January apply to both generations, not only to the new one.

Flash vs Pro, Head to Head

Two chess pieces facing each other on a board in dramatic side light — a deliberate contest, not a race.

Flash versus Pro is the comparison most readers actually need, and it is a comparison of reasoning depth against price, not of features against features. The two models share almost everything on paper: the same 1,048,576-token input context, the same 65,536-token maximum output, the same multimodal input, the same code execution, context caching, file search, function calling, grounding, structured outputs and thinking. What differs is what each model does inside that shared frame — and what it costs. On prompts up to 200,000 tokens, 3.1 Pro costs $2.00 per million input tokens against 3.8 Flash’s $0.75, and $12.00 per million output tokens against Flash’s $3.75. Pro is roughly 2.7 times Flash on input and 3.2 times on output. Against 3.1 Flash-Lite, Pro is eight times the price on both. The question is never whether Pro is better in the abstract. It is whether Pro is better at your task by enough to justify paying three to eight times as much for it.

Reasoning depth is where Pro earns its premium. Google describes Pro as built for multimodal understanding and agentic capabilities, with better thinking and a more grounded, factually consistent experience, optimised for software engineering and for agentic workflows that require precise tool use and reliable multi-step execution. In practice, that shows up in the places where a model has to hold a plan together across many steps: a refactoring that touches a dozen files, an agent that must call the right tool in the right order and notice when a result does not match the plan, a task that mixes a document, an image and a codebase and requires the answer to stay consistent with all three. Flash can do all of these things; Pro does them more reliably, and reliability is the commodity you are buying. A model that gets a ten-step task right nine times out of ten is a different tool from a model that gets it right seven times out of ten, even if their single-step answers look identical.

There are also differences that run in Flash’s favour, and an honest comparison states them. Flash supports computer use; Pro does not list it. Flash lets you set the thinking level — low, medium or high — so you can match the reasoning effort to the task and pay only for the thinking you need; Pro’s thinking is not adjustable in the same way. Flash is a stable model; Pro is still in preview, under the API code gemini-3.1-pro-preview. And Flash has a free tier, while Pro can only be tried free inside AI Studio. None of these differences makes Flash “better” than Pro. They make the two models different tools for different jobs, and they are the reason the default should be Flash, with Pro reserved for the work that justifies it. For readers weighing the same kind of decision in video rather than text — where the model choice is about generation rather than reasoning — our comparison of AI video generators covers the video side of the same question, including the specialist models that the text models in this guide cannot replace.

The decision rule, then, is simple enough to state in one sentence: use Flash by default, use Pro when the cost of a mistake exceeds the cost of the model. That rule has three corollaries worth remembering. First, test both on your actual task before you commit: run the same hard prompt through Flash and Pro in AI Studio, where Pro is free to try, and look at the answers rather than the price list. Second, consider splitting the work: let Flash or Flash-Lite do the bulk of a pipeline — the extraction, the classification, the first draft — and reserve Pro for the step where judgement matters. Third, remember that the price gap is not fixed: Flash doubles on 1 January 2027, and while Pro’s rates are not scheduled to change on the same date, the gap between them narrows from that day onward. A decision that is right in November may need re-checking in January, which is why this guide carries its checked date so prominently.

Pricing, Worked Out: What a Million Tokens Actually Costs

Coins and a calculator-free still life: a jar of coins beside a notebook with blank pages and a pen — cost made tangible, no legible numbers.

Per-million-token prices are abstract until you turn them into a bill, so this section works the examples through. All figures use the Standard tier prices as of 13 November 2026, and all are labelled with the dates on which they change. The arithmetic is simple — input tokens times the input rate, plus output tokens times the output rate — and the only thing that makes it complicated is remembering that thinking tokens count as output, and that Pro’s rate depends on prompt size.

Take a heavy single session: one million input tokens and one hundred thousand output tokens. On Gemini 3.1 Pro, with a prompt of up to 200,000 tokens, that session costs about $2.00 of input plus $1.20 of output — $3.20 in total. On Gemini 3.8 Flash, the same session costs about $0.75 plus $0.38 — $1.13. On Gemini 3.1 Flash-Lite, it costs about $0.25 plus $0.15 — $0.40. On Gemini 4 Carbon, through 30 November 2026, it costs about $2.00 plus $1.00 — $3.00. Four models, the same session, and the bill runs from forty cents to three dollars and twenty cents. That is the whole pricing story in miniature: the model you choose matters far more than the number of tokens you use, and the cheapest model costs one-eighth of the flagship for the same work.

Scale it up to everyday production volume — ten million input tokens and one million output tokens — and the pattern holds. Gemini 3.1 Pro costs about $20 plus $12, or $32, for prompts up to 200,000 tokens. Gemini 3.8 Flash costs about $7.50 plus $3.75, or $11.25. Gemini 3.1 Flash-Lite costs about $2.50 plus $1.50, or $4.00. At this scale the choice of model is a budget decision, not a technical one, and the right question is whether the extra $28 between Flash-Lite and Pro buys $28 of better outcomes. For a pipeline that extracts structured data from documents, it almost certainly does not. For a pipeline that writes production code, it very well might. The per-million-token comparison makes the same point from the other direction: Pro costs about 2.7 times Flash on input and about 3.2 times on output, and eight times Flash-Lite on both. Those ratios are the numbers to carry in your head; the absolute dollars follow from your volume.

There are three ways to cut any of those bills without changing models. The first is the Batch API, which costs 50% of the Standard rate on the models in this guide — so the heavy session above on 3.8 Flash drops from about $1.13 to about $0.56, and 3.1 Pro drops to $1.00 in and $6.00 out per million tokens for prompts up to 200,000 tokens. Google states the 50% cost reduction; this guide states nothing about Batch turnaround times, because Google’s pages read for this guide do not publish them. The second is Flex pricing, which equals Batch pricing for most models — again, a 50% reduction against Standard. The third is Priority pricing, which costs about 1.8 times the Standard rate and is the option for workloads that need guaranteed capacity rather than the lowest price. Between them, Batch and Flex are the single biggest lever on a Gemini bill: halving the rate matters more than any model switch except the jump from Pro to Flash-Lite.

Two more charges belong in any honest cost estimate. The first is thinking: because thinking tokens are billed as output tokens, a reasoning-heavy model on a hard task costs more than its final answer suggests. A model that spends fifty thousand tokens thinking before writing a five-thousand-token answer has billed you for fifty-five thousand output tokens, not five thousand. That is not a hidden charge — Google states it plainly — but it is easy to forget when you are estimating from the length of the answers you see. The second is grounding. Google gives 5,000 free Google Search grounding requests a month, shared across all Gemini 3 and newer models, and then charges $14 per 1,000 requests, with a daily cap of 1,500 requests. Grounding with Google Maps works the same way: 5,000 free requests a month, then $14 per 1,000. If your application grounds every answer in a live search, those requests are a real line on the bill, and the free allowance is shared across models rather than per model — so a system that grounds heavily on Flash and Pro together draws from the same 5,000. For readers interested in how AI answers cite sources in the first place, our guide to AEO, GEO and SEO explains how AI-generated answers choose what to cite, and our deeper guide to GEO covers how content earns those citations — the same grounding pipeline, seen from the publisher’s side.

One final caution about prices, and it is the most important one in this guide. Four price changes land within about seven weeks of publication: Gemini 4 Carbon and Argon double on 1 December 2026, Gemini 3.8 and 3.6 Flash double on 1 January 2027, and the 2.5 auto-routing takes effect on 1 December 2026. Any cost model built on the rates in this guide needs those dates attached to it, and any estimate that quotes a Gemini price without a date is already out of date or about to be. The checked date at the top of this guide — 13 November 2026 — is not decoration. It is the date the prices were true, and the final section of this guide explains how to re-check them.

Context Windows: What 1 Million Tokens Really Means

An open book whose pages stretch impossibly long across a table into the distance, warm library light.

Every current Gemini text model has a 1-million-token input context window. Say it again, because it is the most misunderstood fact in this guide: every one of them. Gemini 4 Carbon, Gemini 4 Argon, Gemini 4 Hydrogen, Gemini 3.1 Pro, Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.1 Flash-Lite all accept one million input tokens. The maximum output is 65,536 tokens for all of them, except that the Gemini 4 models can generate up to 1 million tokens in a single long decode. Context window is therefore not a differentiator between these models. If a roundup tells you to pick one Gemini model over another because of its context window, it is describing an older lineup or a different model family. Within the current lineup, the context is the same, and the choice turns on reasoning depth, speed and price.

What a 1-million-token context window does is set the size of the working memory you can give the model in a single request: the documents, the codebase, the conversation history, the instructions, all of it together, up to a million tokens. That is a very large working memory, and it is the same size whether you are paying $0.25 per million tokens or $2.00. The practical consequence is that the expensive models do not buy you more room; they buy you better use of the same room. A cheap model with a million tokens of context can lose track of the task halfway through a long document; an expensive model with the same million tokens is more likely to stay consistent, notice contradictions and keep the plan in mind. The window is the same. The attention inside it is what you are paying for.

There is one more distinction that causes real confusion, and it is worth stating plainly: the API context window and the app context window are different things. The 1-million-token figure in this guide is the API context window — the size of a request you can send through the Gemini API or Google AI Studio. The Gemini app, the consumer product, has its own context limits, and Google’s plans page confirms an expanded token context window of 1 million tokens for Google AI Pro and Ultra subscribers. What the plans page does not confirm, and what this guide therefore does not state, is the context window for free-tier or Google AI Plus users of the app. Do not conflate the two numbers. A developer sending a million tokens through the API and a subscriber chatting in the app are using different products with different limits, even when the same model name appears in both places. For the concepts underneath all of this — what a context window is, how models hold information, and how techniques like retrieval fit in — our explainer of BERT, LLaMA, Whisper and RAG covers the foundations in plain language.

The output side deserves its own note, because it is where the Gemini 4 models genuinely differ. Every 3.x model caps its output at 65,536 tokens in a single response. The Gemini 4 models list a maximum output of 65K to 1M tokens — long-decode generation up to a million tokens. That is a capability for a specific kind of work: generating very long outputs in a single pass, such as a large codebase, a long report, or a multi-part document that must stay coherent from beginning to end. For most work it will not matter — few answers need more than 65,536 tokens — but for the work that does, it is the difference between one coherent generation and a series of stitched-together ones. If your workload regularly hits the 65,536-token output ceiling, Gemini 4’s long decode is a reason to look at the new generation. If it does not, the output limit is not a reason to choose between models at all.

Google AI Ultra Is a Plan, Not a Model — What It Actually Unlocks

A premium membership card and key on a dark velvet surface beside an open door glowing with light — access, not a product.

This is the section the queue title got wrong, and it is the section this guide exists to get right. Google AI Ultra is not a model. There is no model named Gemini Ultra, Gemini 3 Ultra or Gemini 4 Ultra anywhere in Google’s model catalogue, its pricing page or its plans page. A search for “Ultra” in the model catalogue returns only two hits, neither of them a model: the descriptive phrase “ultra-low latency” attached to an image model, and a shut-down text-to-image model described as having “fast and ultra-fast generation”. What Google’s plans page actually says, in substance, is that there are three Google AI plans that users can subscribe to: Google AI Plus, Google AI Pro and Google AI Ultra. Ultra is the top subscription tier. It unlocks models; it is not one of them. Treating it as a third model alongside Flash and Pro would be comparing a subscription with the things the subscription buys, and this guide does not do that.

What Ultra costs is straightforward, and it is not the price most roundups still quote. Google AI Ultra costs $99.99 a month, billed monthly only, as shown on Google’s plans page savings calculator. The higher launch price from May 2026 that still circulates in older articles is out of date; the current price is $99.99, and this guide states only the current figure. Two tiers exist within Ultra: Ultra 5x, with 20 TB of storage and five times the Gemini usage limits of Pro, and Ultra 20x, with 30 TB of storage and twenty times the limits. Which of those tiers the $99.99 monthly price buys was not determinable from the plans page as fetched for this guide, and this guide does not state a price for the 20x tier rather than guessing one. That is a gap in the public information, stated honestly, not a number invented to fill it.

What Ultra unlocks is the reason it exists. The headline benefit is the highest access to Gemini 4 Carbon and Gemini 4 Argon — the two newest reasoning models — along with higher access to Gemini 3.1 Pro, Deep Research and Deep Think, the deep reasoning mode. It also includes Gemini Agent, available in the US only and in English only, and Project Genie. Beyond the models, the bundle includes a YouTube Premium individual plan, 20 to 30 TB of storage depending on tier, $40 to $100 a month in Google Cloud credits, and 1,000 to 2,000 Colab compute units a month. It is available in more than 150 countries. In short, Ultra is a bundle for a heavy user: the newest models at the highest limits, plus storage, credits and services that would cost real money separately.

Who should pay for it, and who should skip it, follows from that description. Ultra is for the person who uses Gemini heavily every day, who needs the newest reasoning models at the highest limits, who will use the storage and the credits, and for whom $99.99 a month is less than the value of the time it saves. It is not for the person who chats occasionally, who would be served by the free tier or by Google AI Pro at $19.99 a month, or who wants a single model rather than a bundle. The comparison to make is not “Ultra versus Flash” — that question has no meaning — but “Ultra versus Pro”: does the step from $19.99 to $99.99 buy you models, limits, storage and credits you will actually use? For most people, the honest answer is no, and Pro is the better buy. For the heavy user who lives in Gemini, Deep Research and Deep Think all day, the answer may be yes. The same “which tier is worth paying for” question arises with every AI subscription, and readers weighing it across vendors will find our comparison of ChatGPT Free, Plus and Pro useful as the parallel decision on the other side of the market. And for the Search side of the same question — what Pro and Ultra unlock inside Google Search and AI Mode, including expanded and higher access to Gemini 3 Pro in AI Mode features such as Deep Search — our guide to Google AI search features covers the details. Note the wording there carefully: Google’s plans page says “Gemini 3 Pro” for Search, not 3.1 Pro, and this guide quotes it as written rather than correcting it.

Where Each Model Lives: Gemini App vs AI Studio vs the API

Three connected workspaces — a phone, a laptop with a code editor, and a server rack — linked by a single glowing line.

The same model name can mean different things in different places, because Gemini reaches you through three doors: the Gemini app, Google AI Studio, and the Gemini API. The models are the same, but the access, the price and the limits are not, and the right door depends on what you are doing. A casual user will never touch the API. A developer building a product will live in it. A person evaluating a model before committing will find AI Studio the most useful of the three. This section maps each model to each door, and says plainly where the public information runs out.

The Gemini app is the consumer door: the chatbot on your phone and in your browser. Access to models inside the app is governed by your subscription plan, and the plans page describes the tiers as follows. The free tier gives access to the Gemini app, though the exact model and limits for free users are not stated on the pages read for this guide, and this guide does not state them. Google AI Plus gives access to Gemini 4 Hydrogen and to Deep Research, with 400 GB of storage; its price was not shown on the plans page as fetched, so this guide does not state a Plus price rather than repeating the figure quoted by third-party sources. Google AI Pro, at $19.99 a month or $199.99 a year, gives access to Gemini 4 Argon for frontier performance, expanded access to Gemini 3.1 Pro and Deep Research, Gemini inside Gmail, Docs and Sheets, Tasks in Gemini, Chrome auto browse, 5 TB of storage, a YouTube Premium Lite plan, 200 Colab compute units a month, and $10 a month in Google Cloud credits. Pro is also free for one year for students, according to the plans page. Google AI Ultra, at $99.99 a month billed monthly only, gives the highest access to Gemini 4 Carbon and Gemini 4 Argon, higher access to Gemini 3.1 Pro, Deep Research and Deep Think, Gemini Agent in the US only and in English only, Project Genie, 20 to 30 TB of storage, a YouTube Premium individual plan, $40 to $100 a month in Google Cloud credits, and 1,000 to 2,000 Colab units a month. Pro and Ultra subscribers also get the expanded 1-million-token context window in the app. For readers who use Gemini inside Google’s productivity apps, our guide to Google AI in Workspace explains where Gemini Pro and Flash show up inside Docs, Sheets, Slides and Gmail, and what they can do there.

Google AI Studio is the testing door, and it is the most useful one for anyone choosing between models. It is where you can try Gemini 3.1 Pro free, even though the model has no free API tier — the distinction matters, because “no free tier” does not mean “no free trial”. AI Studio is where a developer should run the same hard task through Flash and Pro side by side before committing to a paid plan, and where a product team should prototype before writing production code. The Gemini API is the production door: every model in the table in section 2 is available through it, with a free tier where one exists and pay-as-you-go billing otherwise. Vertex AI is the enterprise route Google presents alongside the Gemini API, and this guide mentions it only in that general sense: per-model Vertex availability was not verified on the pages read for this guide, so no claim is made here about which models are available on Vertex and which are not. That is a limit of this guide, stated rather than papered over, and a reader who needs Vertex specifically should check Google’s Vertex documentation for the model in question.

The practical advice that follows from the three doors is simple. If you are a regular user, live in the app and choose your plan by how much you use it — free for occasional use, Pro at $19.99 for heavy use, Ultra only if you genuinely need the highest limits and the newest models every day. If you are a developer, prototype in AI Studio, where Pro is free to try, and move to the API for production, where the free tier covers the Flash family and Hydrogen and the paid tiers cover the rest. And if you are building a system that will call the API thousands of times a day, design the system around the model ladder in this guide — Flash-Lite for volume, Flash for the default, Pro and Carbon for the hard steps — rather than around a single model for everything. The doors are not interchangeable, and the cheapest way to use Gemini is usually to use the right door for each kind of work.

The Models Google Has Retired: 2.5 and the Original Gemini 3 Pro

An old, retired machine covered with a cloth in a workshop corner, soft dust motes in a beam of light.

Google retires models, and it does so on a schedule that is easy to miss if you are not watching for it. Two retirements matter for anyone with existing code or older tutorials. The first is the original Gemini 3 Pro, API code gemini-3-pro-preview. It no longer appears in Google’s current model list at all. The current Pro is Gemini 3.1 Pro, and the original 3 Pro is gone from the catalogue. Third-party reports say the original was shut down in March 2026 and that its model ID now points to 3.1 Pro; this guide states only what Google’s own pages support — that the original 3 Pro no longer appears in the model list and that 3.1 Pro is the current Pro — and treats the third-party account as consistent with that, not as a verified fact of its own. If your code still names gemini-3-pro-preview, check what it actually returns today, and plan to move it to gemini-3.1-pro-preview.

The second retirement is bigger, because it affects the models many tutorials still teach. Gemini 2.5 Flash and Gemini 2.5 Pro are deprecated. They are no longer selectable in AI Studio, and API access is limited to existing users. From 1 December 2026, all API requests to those models are automatically re-routed: requests to 2.5 Flash go to Gemini 3.5 Flash-Lite, and requests to 2.5 Pro go to Gemini 3.8 Flash, billed at the target model’s rates. That re-routing is automatic, which means old code will keep working — but it will be running a different model than the one its author chose, at a different price, with different capabilities. The honest response is to migrate deliberately rather than to let the automatic routing decide for you. If your code names 2.5 Pro, move it to 3.8 Flash yourself, test the results, and confirm the behaviour you depend on is still there. If it names 2.5 Flash, move it to 3.5 Flash-Lite the same way. One member of the 2.5 family is not deprecated: Gemini 2.5 Flash-Lite continues to be served until further notice, though Google directs new projects to Gemini 4 Hydrogen or 3.5 Flash-Lite instead. And in the video world, a smaller routing change takes effect on 3 December 2026, when Gemini Omni 1.1 Flash routes to Omni 1.2 Flash — a detail for video workloads, mentioned here only so that it is not missed.

There is a lesson in the way Google handles these retirements that applies beyond Gemini. The company keeps older models available for a time, labels them clearly, publishes the re-routing dates, and then moves traffic automatically. That is a gentler retirement than a hard shutdown, but it is still a retirement, and the automatic routing is not a substitute for a deliberate migration. The checklist for any team running Gemini in production is short: search your codebase for gemini-3-pro-preview, gemini-2.5-flash and gemini-2.5-pro; replace each with the current model this guide names; test the replacement on your real workload; and note the 1 December 2026 date in your calendar, because that is the day the automatic routing starts making the decision for you. The same discipline applies to documentation: if a tutorial you rely on names a retired model, treat the rest of its advice with the same caution, because a guide that has not been updated for the retirement has probably not been updated for the price changes either.

How to Pick: A Decision Guide by Task — and Keeping This Guide Current

A signpost at a crossroads in a misty landscape, several paths leading in different directions.

The decision guide below is the whole article compressed into five lines. Each line names a task, the model to use, and the reason — and each reason traces back to a fact established earlier in this guide, not to a feeling about the model names.

Everyday chat, drafting and summarising. Use Gemini 4 Hydrogen or Gemini 3.8 Flash, in the app or through the free tier. The task is simple, the failure cost is low, and the cheapest capable model is the right one. If you are paying for a subscription, Google AI Plus gives you Hydrogen and Deep Research; if you are not, the free tier covers the same kind of work.

Coding, especially long or multi-file work. Use Gemini 3.1 Pro, or Gemini 4 Carbon if you have Ultra access and the task is the hardest kind of reasoning work. Pro is optimised for software engineering behaviour and for agentic workflows that need precise tool use across many steps, and it is the model whose premium is most often justified by the work. Test both Pro and Flash on your actual codebase in AI Studio before you commit, and remember that Carbon’s price doubles on 1 December 2026.

Agents that act on your behalf. Use Gemini 3.8 Flash as the default and 3.1 Pro where reliability matters most, with Flash-Lite handling the subagent steps. Flash is the most intelligent Flash model, supports computer use, and lets you set the thinking level; Pro adds the reliability for the steps where a wrong tool call is expensive. Let Flash-Lite classify and triage, as Gemini CLI does, and spend the Pro budget only where it earns its keep.

Bulk processing: translation, extraction, document parsing, thousands of calls a day. Use Flash-Lite — 3.1 for the lowest price, 3.5 for document parsing and subagent work. At $0.25 per million input tokens, the cheapest model in the family turns a workload that would cost hundreds of dollars on Pro into one that costs tens, and for simple, repeatable tasks the output quality is close enough that the savings dominate. Run the Batch API as well and halve the bill again.

Research that needs grounding in live sources. Use any current model with grounding enabled, and budget for it: 5,000 free Google Search grounding requests a month, shared across all Gemini 3 and newer models, then $14 per 1,000 requests, with a daily cap of 1,500. The model choice matters less here than the grounding itself, so pick by price and spend the savings on the grounding calls.

Two things remain: the choice of plan, and the maintenance of this guide. On plans, the decision is simpler than the model decision. Google AI Pro at $19.99 a month is the right subscription for a heavy user who wants Argon, expanded Pro access and the Workspace integration; Google AI Ultra at $99.99 a month is for the user who needs the highest access to Carbon and Argon, the 5x or 20x usage limits, and the storage and credits that come with it. Most people should not pay for Ultra, and the fact that it is the most expensive option is not a reason to buy it. For the model-picker method itself — the framework for choosing between AI models in general, beyond Gemini — our guide to how to choose an AI model sets out the decision process this article applies. And for the wider field, the natural next read is our explainer of the OpenAI model family, which covers the other side of the market without pretending this guide has compared the two — a genuine Gemini-versus-OpenAI comparison is a different article, and this one does not claim to be it.

On maintenance: this guide was checked on 13 November 2026, and four dates in it are already scheduled to change the facts. Gemini 4 Carbon and Argon double in price on 1 December 2026. Gemini 2.5 requests start auto-routing on 1 December 2026. Gemini 3.8 and 3.6 Flash double in price on 1 January 2027. And Gemini Omni 1.1 Flash routes to Omni 1.2 Flash on 3 December 2026. Any reader who reaches this guide after those dates should re-check the prices on Google’s pricing page and the lineup on Google’s model catalogue before spending money on the strength of it. The links are in the Sources section below, and they are the pages this guide was built from. A model guide is a snapshot, not a permanent truth, and the honest ones say so on the page. This one does.

Sources

Prices and model details checked 13 November 2026 against Google’s official documentation, pricing page and plans page. All facts in this guide come from the sources below, read in full on that date. No third-party roundup was used as a source for any price, limit or capability; where third-party pages disagreed with Google’s own pages, Google’s pages were followed.

  • Google model catalogue — https://ai.google.dev/gemini-api/docs/models
  • Google API pricing page — https://ai.google.dev/pricing
  • Gemini 3.1 Pro model page — https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview
  • Gemini 3.8 Flash model page — https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash
  • Gemini 3.6 Flash model page — https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash
  • Gemini 3.5 Flash-Lite model page — https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite
  • Gemini 3.1 Flash-Lite model page — https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite
  • Google AI plans page — https://one.google.com/about/google-ai-plans/
  • Gemini 3.1 Pro model card, Google DeepMind — https://deepmind.google/models/model-cards/gemini-3-1-pro/

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.