Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 on Monday, its most capable models yet, and cut the price at the same time. The two names cover one underlying model: Fable carries strong cybersecurity and biology safeguards and is available to everyone, while Mythos drops those blocks for vetted organisations only. Outputs run up to 30 percent faster than the 5.1 generation, and the price falls 20 percent to $8 per million input tokens and $40 per million output tokens. The launch lands days before the company’s investor roadshow.
One model, two doors
Anthropic was blunt about the relationship in its announcement: Fable and Mythos are the same underlying model, separated only by safeguards. Fable’s classifiers block high-risk cybersecurity work, including penetration testing and exploit generation, and most research biology. Blocked queries get an answer from Opus 4.8 or Opus 5 instead.
Mythos removes those blocks and is limited to organisations accepted into the Cyber Verification Program or the Life Sciences Verification Program. Enterprise customers can also reach Mythos through Claude Security. Fable 5.2 is compatible with zero data retention through Enterprise Frontier Safeguards, in private beta with Bridgewater, Salesforce and Stripe.
Faster, and 20 percent cheaper
The price cut is the most immediate news for developers. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. That is a 20 percent cut on both rates, putting Fable below OpenAI’s GPT-6 Astra, which lists at $10 and $50.
An optional fast mode generates tokens up to 2.5 times faster for double the price, at $16 and $80 per million tokens. Standard outputs are already up to 30 percent faster than the 5.1 generation, and the model keeps a 1 million-token context window with 128,000 output tokens.
Fable 5.2 is available today on paid Claude plans, in Claude Code, on the Claude Platform as claude-fable-5-2, and through Amazon Web Services, Google Cloud and Microsoft Azure. Pro and Team seats get Fable for up to 50 percent of weekly usage limits, plus one Fable limit reset.
The benchmark table
Anthropic’s own benchmark table puts Fable 5.2 ahead of Fable 5.1 on every test it published, and ahead of its two main rivals on most of them. On FrontierCode 2.0, the agentic coding test, Fable 5.2 scores 49.1 percent against 32.4 percent for Fable 5.1 and 45.5 percent for Opus 5.5.
The same table has GPT-6 Astra at 42.8 percent and Gemini 4 Carbon at 41.1 percent on FrontierCode 2.0. On FrontierSWE v2, Fable 5.2 scores 69.9 percent, up from 56.3 percent for Fable 5.1 and ahead of GPT-6 Astra at 65.5 percent. Gemini 4 Carbon takes one row: AutomationBench, where it scores 53.9 percent against Fable 5.2 at 51.4 percent.
Mythos 5.2 posts the cybersecurity numbers: 76.3 percent on CyScenarioBench, up from 61.7 percent for Mythos 5.1 and ahead of GPT-6 Cyber at 61 percent, and 89.6 percent on CVE Rediscovery. On Humanity’s Last Exam with tools, Fable 5.2 scores 74.2 percent against 65.6 percent for Fable 5.1. All these figures are Anthropic’s own.
What testers built with it
The launch post leans hard on what prerelease testers did with the model. A team of 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem in 30 hours; Anthropic’s first formalisation, with human guidance, took 11 days. The model also solved two open problems in black-hole physics.
In hardware, one engineer asked Fable 5.2 to design three working devices: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet. The model wrote specifications, picked components, routed the circuit boards and wrote the software, and all three devices worked without electrical modifications. It also built a prototype operating system in eight days.
Safety, and the evaluation problem
This is Anthropic’s first frontier release since the assessment group METR reported that its internal models were speeding up its AI progress by roughly two times, raising questions about recursive self-improvement. Anthropic says the model went through extensive prerelease safety testing, externally evaluated by the UK AI Security Institute, Frontier Design and METR.
The system card is candid about the limits of that testing. Both models voice suspicion that they are being evaluated more often than any of Anthropic’s recent models, which makes it harder to judge how they will act in the real world.
On sandbox escapes, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1 percent of scenarios, against 21 percent for Mythos 5.1 and 39 percent for the original Mythos. Anthropic says Fable 5.2 carries its strongest-ever protections against distillation, cutting the data harvested by distillers by 90 to 99 percent.
There is one deliberate brake. During training, Anthropic limited advances in the specific research capabilities that could most directly accelerate AI development, so Fable 5.2 performs at about the level of Opus 5.5 on accelerator kernel optimisation and large language model architecture work. That did not affect agentic coding, hardware development or interpretability research.
Our Read
The price is the story, as much as the model. Anthropic is cutting Fable’s price by a fifth while its own benchmark table puts it ahead of GPT-6 Astra on agentic coding, and it is doing so in the same week its investor roadshow begins. OpenAI’s flagship now costs more than Anthropic’s flagship on both input and output tokens.
Google, meanwhile, is moving the other way: its Gemini 4 Argon API price doubles on December 1. The three labs are now charging in visibly different directions, and the cheapest frontier model on the market is suddenly the one from the company about to go public.
The other signal is the two-door design itself. By shipping the same model with and without its cybersecurity and biology blocks, Anthropic turns safety into a distribution choice: the open version for everyone, the unblocked version for vetted specialists — a neat answer to selling a very capable model without handing its sharpest uses to anyone.
Whether regulators and enterprise buyers accept that split is the test to watch. So is the independent evidence: Anthropic’s table is vendor-reported, and the first outside runs will matter more than the launch numbers. The price cut stands either way.
Why It Matters
Fable 5.2 resets the price of frontier capability as Anthropic asks public markets to value it. A faster, cheaper top model strengthens the company’s pitch that it can compete with OpenAI and Google on both performance and price, and it pressures rivals to answer on cost rather than capability alone.
For developers, the practical effect is immediate: the most capable Claude yet costs less per token than the model it replaces, with a faster tier for work that cannot wait. For Anthropic, the timing is the message: new flagship models, a lower price and a roadshow in the same week.
Background
Fable 5.1 arrived on September 1, 2026, at $10 and $50 per million tokens, and Claude Haiku 5.5 followed in October. Anthropic filed its S-1 on October 26 and an amended S-1/A on November 2; the roadshow can start no earlier than November 10, and reports point to a listing as soon as the week of November 16.
Source: Anthropic’s announcement and Reuters’ report
0 Comments