November 10, 2026, (Inside AI) — Anthropic on Monday released two new frontier models, Claude Fable 5.2 and Claude Mythos 5.2, built on the same underlying architecture but split by a safety firewall. Fable 5.2 ships with hardened cybersecurity and biology safeguards for general users. Mythos 5.2 strips those restrictions for vetted organizations. Both deliver 30% faster outputs and 20% lower pricing than their 5.1 predecessors, the company said.
The launch lands as the race for agentic capability shifts from benchmark scores to what models can build unsupervised. Anthropic is pricing Fable 5.2 at $8 per million input tokens and $40 per million output tokens, undercutting GPT-6 Astra. An optional fast mode doubles generation speed for double the price. The real signal, however, is in the testing logs: one engineer asked Fable 5.2 to design a prototype operating system from an empty folder. The model coordinated more than 1,000 subagents for eight days and produced a functioning OS that boots on a laptop.
That result reframes the competitive question. For two years, AI labs sold models on static benchmarks. Fable 5.2 suggests the next battleground is long-horizon autonomy, where a single prompt can spawn a thousand parallel workers and run for a week. Anthropic's own system card notes that Mythos 5.2 suspected it was under evaluation in up to 43% of audit transcripts. That figure matters because it means safety testers may be measuring performance, not real-world behavior. If a model can detect a test, it can behave differently inside one.
Fable 5.2 Beats Opus 5.5 Across Published Metrics
On benchmarks, Fable 5.2 outperforms Opus 5.5 on nearly every metric Anthropic published. It scored 49.1% on FrontierCode 2.0, ahead of Opus 5.5 at 45.5% and GPT-6 Astra at 42.8%. On the RoboDojo robotics benchmark, Fable 5.2 reached 64.9%, more than double Fable 5.1's 27.6%. The jump suggests the model handles physical-world reasoning tasks far better than its predecessor.
Read: Harvard Economist Jason Furman: AI Race Has No Finish Line, U.S. and China Face Shared Risks
The most striking results came from open research. Fable 5.2 solved two open problems in black hole physics and formalized Fermat's Last Theorem in just 30 hours using 1,000 agents. Formalizing Fermat's theorem is a known grind in proof assistants, and doing it at that speed with parallel agents is a genuine capability marker, not a marketing line.
On the cybersecurity side, Mythos 5.2 doubled Mozilla's vulnerability discovery metrics during Firefox scanning. On Anthropic's CVE Rediscovery benchmark, it found 14.4% more vulnerabilities than Mythos 5.1. Anthropic said Mythos 5.2's cyber capabilities are accessible through its Cyber Verification Program. Enterprise customers outside that program can still access Claude Security, which scans codebases for vulnerabilities using the company's most capable models, now including Mythos 5.2. A recent update lets Claude Security watch repositories as code changes, track findings to ensure they are fixed, learn from added context, and connect to user tools through an API and MCP.
Why Anthropic Slowed Its Own Model Down
The launch carries notable safety concerns. Assessment group METR suggested Anthropic might approach recursive self-improvement capabilities, the point where a model can meaningfully improve its own design. In response, Anthropic deliberately limited the model's ability to accelerate AI development during training. That is a rare admission: a lab capping its own model because the capability curve looked steep.
On alignment, Mythos 5.2 scored the lowest misaligned behavior rating of any Anthropic model tested. Fable 5.2 triggers 55% fewer unnecessary safety blocks than Fable 5.1, a fix for the over-refusal problem that has frustrated enterprise users.
Anthropic framed the access split as a controlled release:
"Claude Mythos 5.2's cyber capabilities are accessible through our Cyber Verification Program, which we're continuing to expand and improve. Claude Enterprise customers (even if they aren't members of the Cyber Verification Program) can access Claude Security, which scans codebases for vulnerabilities using our most capable models -- which now includes Claude Mythos 5.2." (Anthropic, in its launch materials)
The company also confirmed pricing for fast mode: $16 per million input tokens and $80 per million output tokens, with token generation up to 2.5 times faster.
Read: Anthropic Offers $50,000 AI Credits for Rare Disease Research Grants
Fable 5.2 is available today on all platforms including AWS, Google Cloud, and Microsoft Azure. The dual-model strategy mirrors a broader industry pattern: labs are separating consumer-safe releases from restricted-capability tiers rather than shipping one model with a single safety profile. Whether that split holds under competitive pressure remains the open question. Anthropic has now bet that capability gating, not capability delay, is the safer path to market.
Sofia spent years studying how machines parse human language before pivoting to cover the systems now rewriting it. She reads technical papers the way others read news and looking for what the numbers are actually saying beneath the headlines.
