Anthropic has launched its most capable models yet. On November 9, the company introduced Claude Fable 5.2 and Claude Mythos 5.2 — two versions of the same underlying model, split by who is allowed to use them and what they are allowed to do. Fable 5.2 is the general-purpose frontier model, available to everyone. Mythos 5.2 is the same model with its cybersecurity and biology safeguards lifted, reserved for vetted organisations. Both are faster than their predecessors and, unusually for a frontier launch, cheaper: standard API pricing falls 20 percent, to $8 per million input tokens and $40 per million output tokens.
The launch lands at a charged moment. Reuters reports that the new models arrive ahead of Anthropic’s planned IPO, and the company is pitching them as its strongest answer yet to OpenAI, Google and Meta in the race for the most demanding technical work — coding, science and security.
One Model, Two Doors: Fable for Everyone, Mythos for Vetted Experts
The most unusual thing about this launch is that Fable 5.2 and Mythos 5.2 are not two different models. According to Anthropic’s official announcement, they share the same underlying model, the same 1-million-token context window and the same 128,000-token maximum output. What differs is the guardrails and the gate.
Fable 5.2 carries strong safeguards that block most cybersecurity and biology requests — penetration testing, exploit generation, binary vulnerability work and most research-biology tasks. Those blocks are what let Anthropic offer it to all users: it is available from launch day on paid Claude plans, in Claude Code and through the Claude Platform, as well as on Amazon Web Services, Google Cloud and Microsoft Azure. Its API identifier is claude-fable-5-2.
Mythos 5.2 is the same model without those blocks. Direct access is limited to organisations accepted into Anthropic’s Cyber Verification Program or Life Sciences Verification Program. Enterprise customers can also reach Mythos through Claude Security, which scans codebases for vulnerabilities using Anthropic’s most capable models. The split is a deliberate design choice: one model, two levels of trust.
Faster and Cheaper: The New Pricing
Speed and price are the two numbers most developers will check first. Anthropic says Fable 5.2 and Mythos 5.2 generate outputs up to 30 percent faster than their 5.1 predecessors, and testers found they communicate more clearly. The price cut is just as direct: $8 per million input tokens and $40 per million output tokens, down from $10 and $50 — a 20 percent reduction, with cache reads falling to $0.20 per million tokens.
There is also a new optional fast mode, which speeds up token generation by up to 2.5 times for $16 per million input tokens and $80 per million output tokens. VentureBeat’s analysis notes an important nuance for buyers: the list price is only part of the bill. On Cognition’s FrontierCode 2.0 test, Fable 5.2 at High effort averaged $6.72 per task — roughly half what Opus 5.5 spent at Max for a nearly identical score — while turning the effort dial to Max pushed the average to $44.70 and actually lowered the score. The effort setting, not just the rate card, decides what a job costs.
The Benchmarks: Coding Leads the Way
Anthropic’s benchmark table puts Fable 5.2 a step above Opus 5.5 in almost every capability it evaluated, and ahead of its named rivals on most lines. The headline number is FrontierCode 2.0, an agentic coding test that asks whether a patch is good enough to merge into a real open-source repository: Fable 5.2 scored 49.1 percent at Xhigh effort, against 32.4 percent for Fable 5.1, 45.5 percent for Opus 5.5, 42.8 percent for GPT-6 Astra and 41.1 percent for Gemini 4 Carbon.
The lead holds across the coding suite. On FrontierSWE v2, Fable 5.2 scored 69.9 percent, ahead of Astra at 65.5 percent and Carbon at 61.2 percent. On the Terminal-Bench-Science research test it reached 74.6 percent (75.3 percent for Mythos 5.2), and on the OSWorld computer-use test, 89.3 percent. Even the knowledge-work test favours it on the rubric: 66.1 percent on AA-Briefcase, against 61.2 percent for Opus 5.5 — although Opus still edges it on the head-to-head Elo ranking, 1,800 to 1,785.
Two caveats belong with these numbers. First, they are Anthropic’s own results, published in its launch announcement and system card, and rival models were run at different effort settings in some rows. Second, the gains are not uniform: Gemini 4 Carbon still leads on the AutomationBench business-workflows test, 53.9 percent to Fable’s 51.4 percent. The honest reading is that Fable 5.2 leads on coding and science, not on everything.
Science Claims: Black Holes, Proteins and Fermat
The most eye-catching part of the announcement is not a benchmark table but a list of things the models did during prerelease testing — and these deserve to be reported as Anthropic’s claims, not settled facts. A team of about 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem in the Lean proof language in about 30 hours, with minimal human intervention. To be clear about what that means: the model did not discover a new proof of the centuries-old theorem. It translated an existing proof into a form a computer can check step by step — a laborious formalisation task that took an earlier Claude setup 11 days. The new attempt needed one-third as many action tokens and produced a proof seven times shorter.
In theoretical physics, Fable 5.2 solved two open problems about black holes, both concerning the effort to unite relativity and quantum mechanics. Anthropic has pointed to preprints for the results, and a preprint is an invitation to scrutiny, not peer-reviewed acceptance. In biology, Mythos 5.2 found a new way to predict the 3D structure of whole viral vector capsids — the protein shells used to deliver gene therapies — accurately modelling 28 of the 34 capsids it attempted.
The hardware claim is the most concrete of the three. Given a one-paragraph prompt, Fable 5.2 designed three working electronic devices from scratch — a home voice-assistant hub, a USB speakerphone and a handheld virtual pet — selecting components, routing four-layer circuit boards, modelling 3D-printed enclosures and writing the software. Each board was ready for manufacturing in under a day, and all three devices worked without electrical modifications once built by partner Diode. Separately, more than 1,000 subagents built a working prototype operating system that boots on a laptop in eight days.
Mythos for Cybersecurity: The Red-Teaming Edge
Mythos 5.2 is Anthropic’s most capable model for cybersecurity engineering, and the numbers explain why the company keeps it behind a verification wall. On its internal CVE Rediscovery evaluation, Mythos 5.2 found 10.2 percent more vulnerabilities than Opus 5.5 and 14.4 percent more than Mythos 5.1, scoring 89.6 percent against 78.3 percent for its predecessor. A security company testing the model found that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model it tested.
The same capability is already showing up in enterprise products. Cisco’s AI Defense now uses Claude Mythos to red team enterprise AI agents, generating multi-step attack paths to surface vulnerabilities before agents reach production. The pairing makes sense: a model that can find flaws in code can also be aimed at the agents built on top of it. Anthropic says Mythos 5.2 also shows better judgment than its predecessors — one tester found earlier models had overstated the seriousness of about 1 in 20 findings, while Mythos 5.2 had not overstated any.
Where to Use It: Fable, Opus and the Model Ladder
Anthropic’s own guidance is refreshingly restrained: for most coding and knowledge work, start with Claude Opus 5.5, described as the company’s strongest daily driver, and move up to Fable or Mythos only for the most ambitious, open-ended technical work — or down to Sonnet 5.5 or Haiku 5.5 for lighter jobs. Fable 5.2 can also play two supporting roles: Advisor, where smaller models consult it on harder problems, or Orchestrator, where it delegates simpler tasks to smaller models.
The ecosystem moved fast. Cognition made Fable 5.2 available in Devin Desktop and Devin CLI on launch day, and Anthropic says more than 90 percent of the changes merged into its own core codebase are now written by Claude with human guidance. One team of agents migrated a critical production service from Python to Rust; the service now starts about 26 times faster. Developers already living in the Claude workflow will feel the upgrade first: Claude Code 2.1.319 shipped just days earlier with improvements for unattended coding runs, and Fable 5.2 is available inside Claude Code from day one.
What It Means for Developers
For developers, the practical question is not whether Fable 5.2 tops a benchmark table but what it changes about how work gets done. Three things stand out. First, the price cut and the speed gain arrive together, which is rare: a frontier model that costs 20 percent less and runs up to 30 percent faster. Second, the effort settings matter more than the headline scores — the same model can cost $2.48 or $14.19 per FrontierCode task depending on the dial, so teams should pilot at Medium before turning anything up. Third, the agentic direction is now the whole story: long-running agent teams that migrate services, formalise proofs and design hardware are the workloads these models are built for, and the same week brought OpenAI’s multi-computer dots, which let one agent work across several machines. The industry is converging on agents that run for hours across real systems, and Fable 5.2 is Anthropic’s bid to power them.
The safety side deserves equal attention. Anthropic says it ran extensive prerelease safety testing, including novel evaluations for sandbox escapes, and believes high-stakes misalignment risks remain low. It also says it has implemented its strongest-ever protections against distillation attacks, and that Fable 5.2’s safeguards are as robust as Fable 5.1’s while producing fewer unnecessary blocks. Whether those claims hold up will be tested in public now that the model is widely available — which is, in the end, the point of releasing it.
Bottom Line
Claude Fable 5.2 and Claude Mythos 5.2 are Anthropic’s most capable models, faster and cheaper than the generation they replace, with their strongest results in coding, science and cybersecurity. The two-door model — general access for Fable, vetted access for Mythos — is the company’s answer to the hardest question in frontier AI: how to ship the most powerful version of a model without handing its riskiest capabilities to everyone. Fable 5.2 is available now on Claude’s paid plans, Claude Code, the Claude Platform and the major clouds. Start with Opus 5.5 for everyday work, reach for Fable when the job is genuinely hard, and treat the science claims as a preview of what agent teams can do — not as proof that the model did it alone.
