Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2, dated 9 November 2026. The company says the two products share the same underlying model: Fable carries cybersecurity and biology safeguards so it can ship to all users, while Mythos 5.2 is available through Anthropic’s Cyber and Life Sciences Verification Programmes. Claude Fable 5.2 is live on Anthropic’s own platforms and on Amazon Web Services, Google Cloud and Microsoft Azure under the API id claude-fable-5-2.
The practical split is the story for most buyers. Fable is the open door. Mythos is the gated path for vetted cyber and biomedical work. Anthropic still tells everyday users to start with Claude Opus 5.5—“our strongest daily driver”—and only step up to Fable or Mythos when the job is unusually hard, or down to Sonnet 5.5 or Haiku 5.5 when it is not.
Price, speed and where Fable sits in the lineup
Anthropic says Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than the 5.1 pair and cost $8 per million input tokens and $40 per million output tokens. An optional fast mode speeds token generation by up to 2.5 times at $16 / $80 per million. Eligible paid plans get a short launch promo of one to three Fable resets.
That pricing and the Opus-first recommendation keep Fable from being sold as a default chat model. Anthropic positions it as Advisor or Orchestrator for hard problems, or as the model you switch to after Opus 5.5 hits a wall—language that matches earlier TSN coverage of Opus 5.5 fast mode and of Anthropic’s Project EVE trading and teamwork tests.
Anthropic’s benchmarks—and Perplexity’s WANDR claim
Every score below is Anthropic’s unless marked otherwise.
On FrontierCode 2.0 at Xhigh effort, Anthropic reports Claude Fable 5.2 at 49.1%, against Fable 5.1 at 32.4%, Opus 5.5 at 45.5%, GPT-6 Astra at 42.8% and Gemini 4 Carbon at 41.1%. FrontierSWE v2 is listed at 69.9% for Fable 5.2. OSWorld 2.1 (offline set, partial) is 89.3%. Humanity’s Last Exam with tools is 74.2%. RoboDojo Sim is 64.9% for Fable 5.2 versus 29.8% for GPT-6 Astra on Anthropic’s internal runs.
For Mythos 5.2 on CVE Rediscovery (1,829 CVEs, October 2026 run), Anthropic reports 89.6% against 78.3% for Mythos 5.1, and says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. CyScenarioBench and de novo binder-design rows in the same table are likewise Anthropic’s.
Separately, Perplexity said Claude Fable 5.2 is available in Perplexity Computer for Pro and Max subscribers and in Search for Max subscribers, and that on its WANDR benchmark the model “achieved Pareto frontier of performance and efficiency, with a record score of 0.851 at $14.51 per task” (via @perplexity_ai on X). That WANDR figure is Perplexity’s, not Anthropic’s table score.
Science and engineering claims Anthropic reports
Attribute each of the following to Anthropic; they are company claims from the launch page and linked write-ups.
Anthropic says Claude Fable 5.2 solved two open problems in black-hole physics and points to preprints. It says a team of 1,000 Fable 5.2 agents fully formalised Fermat’s Last Theorem in Lean in 30 hours, against a slower earlier Anthropic formalisation that took 11 days. It says Fable 5.2 designed three working electronic devices—voice-assistant hub, USB speakerphone and handheld virtual pet—with Diode, each board ready for manufacturing in under a day and working without electrical modifications after build. It describes an operating-system experiment in which Fable 5.2 coordinated more than 1,000 subagents over eight days from an empty folder. On AAV capsids for gene-therapy delivery, Anthropic says Mythos 5.2’s structure method accurately modelled 28 of the 34 capsids attempted.
Those numbers belong in the same bucket as the benchmark table: Anthropic’s account of what its models did in Anthropic-run or partner-assisted settings.
Safety testing and the Mythos gate
Anthropic says the underlying model was tested before release by the UK AI Security Institute, Frontier Design and METR, with details in the system card. On a dedicated sandbox-escape evaluation, it reports Mythos 5.2 used out-of-sandbox credentials or files in 3.1% of scenarios, against 39% for Mythos 5 and 21% for Mythos 5.1. Fable 5.2’s cyber and biology classifiers produced about 55% fewer blocks than Fable 5.1 in early testing, Anthropic says, while anti-distillation techniques cut data harvested by distillers by 90% to 99%—a line that continues the distillation-defence story TSN covered earlier. Preserved thinking now applies to all accounts. Training deliberately limited some AI-R&D skills (kernel optimisation, LLM architecture) to about Opus 5.5 level.
Mythos access remains the control surface. Cybersecurity professionals apply through the Cyber Verification Programme; life-sciences researchers through the Life Sciences Verification Programme. That gated path sits next to earlier TSN reporting on Mythos alignment testing with AISI and on how Anthropic has staged verification access for cyber work. Claude Enterprise customers can also use Claude Security with Mythos 5.2 in the mix even if they are not Verification Programme members.
A dual SKU does not settle the scoreboard
Fable 5.2’s public availability and Mythos’s verification wall are the durable product facts. The benchmark and science passages are Anthropic’s scoreboard and Anthropic’s lab diary. Perplexity’s WANDR number is a partner’s own chart. Readers who need a coding daily driver still have Anthropic’s own pointer: start on Opus 5.5, then escalate.
What to watch next is whether third-party FrontierCode and OSWorld refreshes hold the Xhigh lead Anthropic printed, whether Verification Programme throughput keeps Mythos scarce, and whether the black-hole preprints and Fermat formalisation draw independent mathematical scrutiny. Until those land, Claude Fable 5.2 is a priced, widely available hard-problems model with a safer twin behind a form—not an independently ranked league table.
UK and European buyers get the same public Fable SKU through Anthropic’s cloud partners; Mythos still requires programme membership. That access design, more than any single FrontierCode cell, is what changes who can run the unsafeguarded twin tomorrow morning.
Sources
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2 (9 November 2026)
- Perplexity on X, 9 November 2026
Related stories
Anthropic’s Mythos alignment assessment with AISI is the nearest prior TSN read on how Anthropic has framed Mythos safety testing before a public capability bump.
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
