Glonce

Anthropic releases Claude Fable 5.2 and Mythos 5.2, with Fable first on Artificial Analysis index

modelsThe Glonce editorsThe Decoder🔥78Glonce rating78Impact78Novelty70Relevance100Surprise45Players95Credibility75
Anthropic releases Claude Fable 5.2 and Mythos 5.2, with Fable first on Artificial Analysis index

Anthropic is releasing Claude Fable 5.2 and Claude Mythos 5.2, two flagship models built for the most demanding technical tasks. The Decoder reports that the release comes shortly before the company's IPO and that the models reclaim the performance lead, in some cases by a wide margin. Opus 5.5 stays Anthropic's recommended model for everyday use.

The two models share one foundation but differ in access, as their predecessors did. Fable has safeguards that block risky tasks in cybersecurity and biology and is available to all users. Mythos removes certain of those restrictions and is open only to organizations that apply through Anthropic's Verification Programs for Cybersecurity or Life Sciences. Both cost $8 per million input tokens and $40 per million output tokens, against $10 and $50 for the predecessors. The article calls that a cut only on paper: what users pay depends on token consumption, which varies with tasks, settings and software environments. Anthropic says the models produce output up to 30 percent faster than Fable 5.1 and Mythos 5.1. An optional fast mode for select organizations speeds up token output by up to 2.5 times at twice the cost, and is available only on certain platforms, including Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry.

On the benchmarks, Artificial Analysis puts Fable 5.2 first on its Intelligence Index v5 at the highest compute tier with 65 points. That is five ahead of Gemini 4 Carbon (60, High tier, a model Google has not yet released), six ahead of Claude Opus 5.5 (59, Max tier) and 15 ahead of Claude Fable 5.1, which came out in early September. Fable 5.2 scores best in five of the nine individual tests: 80 percent on HLE-Diamond (expert-level academic questions), 73 percent on Terminal-Bench-Science (independent scientific workflows in a terminal) and 49 percent on FrontierCode 2.0 (useful code changes in real open-source projects). It is weaker in agentic knowledge work and terminal coding. On AA-Briefcase it reaches 1,785 Elo, just behind Claude Sonnet 5.5 (1,813) and within the confidence interval of Opus 5.5 (1,800). On Terminal-Bench 4.0 it scores 60 percent, tied with Opus 5.5, while Kimi K3.1 Pro (67), Gemini 4 Carbon (65) and Sonnet 5.5 (64) rank higher.

Peak quality is expensive. Per-task costs at the highest tier average $26.50, about 2.6 times Opus 5.5 ($10) and roughly double Fable 5.1 ($12.99). At the lower High and Xhigh tiers Fable 5.2 still scores 62 points, ahead of every other model at any tier according to Artificial Analysis, while using far fewer tokens: 98,000 output tokens per task against 173,000 for Opus 5.5 at the Max tier.

On security, Anthropic says Mythos 5.2 finds more vulnerabilities than earlier models and is better at reproducing and fixing them. The public Fable variant blocks most security-related tasks as a result, including penetration tests and exploit code generation. In Anthropic's internal CVE Rediscovery test, a model is given a project's state before a fix and must find published CVE vulnerabilities on its own. In the October 2026 run with 1,829 CVEs, Mythos 5.2 found 10.2 percent more vulnerabilities than Opus 5.5 and 14.4 percent more than Mythos 5.1, and led at every compute tier, Anthropic says. Anthropic also says one security firm reported that one in six confirmed bugs Mythos 5.2 found had been missed by all other frontier models tested.

Anthropic says it deliberately held the models back during training from improving at the capabilities that could most directly accelerate AI development, so-called RSI. Fable 5.2 and Mythos 5.2 perform roughly on par with Opus 5.5 in areas such as optimizing compute kernels for accelerator chips and improving large language model architectures. Agentic coding, hardware development and traditional AI research, including interpretability research, are not affected, per Anthropic. The company adds that Claude now writes over 90 percent of the changes merged into Anthropic's core codebase, under human guidance, and claims the model programmed an operating system from scratch in an internal test.

The headline example is mathematics. About 1,000 Fable 5.2 agents fully formalized the proof of Fermat's Last Theorem in the Lean programming language, so a computer can verify it step by step, Anthropic says. Andrew Wiles presented a proof of over a hundred pages in 1995; the article says its machine verification has now been completed for the first time, and it took 30 hours. An earlier, unpublished research version of Claude needed 11 days for the same task and relied on a purpose-built environment and guidance from a human researcher. The Fable 5.2 agents started with brief instructions, organized their collaboration on their own and used no mathematical references, Anthropic says. The only progress metric was a list of intermediate lemmas. The model used one-third of the action tokens of the first attempt and produced a proof seven times shorter. For comparison, Anthropic ran Opus 5.5 in the same setup with 1,000 agents: it managed just 26 percent of the formalization in 18 hours and made barely any progress through hour 30.

Key facts

  • Claude Fable 5.2 (open to all users, with cybersecurity and biology safeguards) and Mythos 5.2 (restricted to organizations in Anthropic's Verification Programs) share one foundation and cost $8 per million input and $40 per million output tokens, versus $10 and $50 before.
  • Per Artificial Analysis, Fable 5.2 ranks first on Intelligence Index v5 with 65 points, five ahead of Gemini 4 Carbon (60), six ahead of Opus 5.5 (59) and 15 ahead of Fable 5.1; the highest tier averages $26.50 per task, about 2.6 times Opus 5.5.
  • At the lower High and Xhigh tiers Fable 5.2 scores 62 points using 98,000 output tokens per task, versus 173,000 for Opus 5.5 at the Max tier.
  • Anthropic says Mythos 5.2 found 10.2 percent more vulnerabilities than Opus 5.5 and 14.4 percent more than Mythos 5.1 in an October 2026 CVE Rediscovery run with 1,829 CVEs.
  • Anthropic says about 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in Lean in 30 hours, against 11 days for an earlier unpublished research version; Opus 5.5 reached 26 percent in 18 hours.

Why it matters

Per Artificial Analysis, Fable 5.2 takes first place on Intelligence Index v5 with 65 points, a jump of 15 points over Fable 5.1, which shipped only in early September. The Decoder frames the release as Anthropic reclaiming the performance lead shortly before its IPO. Two other threads stand out: Anthropic says it deliberately held back the capabilities that could most directly accelerate AI development, and it presents the Lean formalization of Fermat's Last Theorem as a sign of what agent teams can now do.

Who it affects

Fable 5.2 is available to all users, with safeguards that block risky cybersecurity and biology tasks. Mythos 5.2 is open only to organizations accepted into Anthropic's Verification Programs for Cybersecurity or Life Sciences. The fast mode is for select organizations on certain platforms, including Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry. Everyday users are pointed to Opus 5.5, which Anthropic continues to recommend.

How to use it

Pricing for both models is $8 per million input tokens and $40 per million output tokens. Anthropic positions them for the most demanding technical work and keeps recommending Opus 5.5 for everyday use. The cost trade-off is clear in the benchmark data: the highest tier averages $26.50 per task, while the lower High and Xhigh tiers still score 62 points with 98,000 output tokens per task. Fast mode speeds up token output by up to 2.5 times at twice the cost.

How solid is it

The index and benchmark results come from Artificial Analysis, not from Anthropic. The cybersecurity figures (10.2 and 14.4 percent more vulnerabilities, the one-in-six figure relayed from a security firm), the speed gain, the 90 percent codebase share, the operating system test and the held-back training are all Anthropic's own statements as relayed by The Decoder. The Fermat result is Anthropic's own account of its run. The source text is garbled in places, so this retelling leaves out those passages.

Risks and caveats

The lower price is a list-price cut only; actual spend depends on token use, which varies with tasks, settings and environments, and peak-tier runs cost about 2.6 times Opus 5.5 per task. Fable 5.2 is not first everywhere: it trails Sonnet 5.5 narrowly on AA-Briefcase and ranks behind Kimi K3.1 Pro, Gemini 4 Carbon and Sonnet 5.5 on Terminal-Bench 4.0. The Gemini 4 Carbon score refers to a model Google has not yet released. No IPO date or valuation is given in the article.