Anthropic releases Claude Fable 5.2 and Mythos 5.2 at lower prices ahead of its IPO
Anthropic's upgraded top models run up to 30% faster and cost less, and the company says 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in about 30 hours. Mythos 5.2 stays limited to vetted security and life-sciences groups.
Anthropic on Monday released Claude Fable 5.2 and Claude Mythos 5.2, an upgrade to its top-tier models that the company says runs up to 30% faster than the 5.1 versions and costs less to use. Fable 5.2 is generally available on the Claude Platform and through Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 stays restricted to organizations vetted through Anthropic's cybersecurity and life-sciences verification programs.
The launch lands days before an expected public listing. CNBC reported that the IPO could take place as soon as next week, and that it is the first frontier release since chief executive Dario Amodei called in September for a slower pace of AI development. "We called for pacing, not pausing," Dianne Penn, Anthropic's head of product management for research, told CNBC.
One model, two sets of guardrails
Fable and Mythos are the same underlying model. The difference is in the safeguards. Fable ships with classifiers that block high-risk cyber work, such as exploit generation and penetration testing, along with most research-biology tasks; blocked queries are answered by an older Opus model instead. Mythos drops some of those blocks and is offered only to approved security and biomedical groups. Anthropic says that in early testing Fable 5.2 triggered about 55% fewer blocks than Fable 5.1, roughly 60% fewer in cybersecurity and 40% fewer in biology.
Pricing is $8 per million input tokens and $40 per million output tokens, which CNBC says is 20% below the 5.1 generation. An optional fast mode, which Anthropic says speeds up token generation by up to 2.5 times, costs $16 and $80. The company still points most users to Claude Opus 5.5 for everyday coding and knowledge work and positions Fable as the model for the hardest, open-ended jobs, or as an "advisor" that smaller Claude models consult.
What the benchmarks show
On Anthropic's own benchmark table, Fable 5.2 scores 49.1% on the FrontierCode 2.0 agentic coding test at its highest effort setting, against 45.5% for Opus 5.5, 42.8% for OpenAI's GPT-6 Astra and 41.1% for Google's Gemini 4 Carbon. The largest jump is in robotics: on the RoboDojo simulation, which tests control of a two-armed robot from camera images, Fable 5.2 scores 64.9%, against 27.6% for Fable 5.1 and 29.8% for GPT-6 Astra, based on Anthropic's internal runs. It is not ahead everywhere. Gemini 4 Carbon scored 53.9% on the AutomationBench business-workflow test, above Fable 5.2's 51.4%.
The company also says the model is more efficient. At its "High" effort setting, Fable 5.2 roughly matched Opus 5.5's best FrontierCode score for about half the cost per task.
On security, Anthropic says Mythos 5.2 found 14.4% more vulnerabilities than Mythos 5.1 on its internal CVE Rediscovery test, which ran against 1,829 published flaws in October. One tester that has spent the year hardening Firefox with successive Claude models told Anthropic that Mythos 5.2 roughly doubled the discovery numbers of Mythos 5.1 across its scanning campaigns. OpenAI has been pushing in the same direction with GPT-6 Cyber.
Fermat in 30 hours
The most striking demonstration is in mathematics. Anthropic says a team of about 1,000 Fable 5.2 agents translated the proof of Fermat's Last Theorem into Lean, a language in which a computer checks every step, in about 30 hours with minimal human help. The company's first formalization in September took 11 days and relied on a purpose-built platform plus guidance from a researcher. According to a technical write-up on Anthropic's developer blog, the agents worked in sandboxes without internet access, every shell command was screened by a classifier, and the team was given 60 target statements and 47 previously proved lemmas but no instructions on how to organize. The resulting proof was about one-seventh the length of the original. Opus 5.5, run in the same setup, completed 26% of the work in 18 hours.
Anthropic lists other prerelease results: two solved open problems in black-hole physics, now posted as preprints; three working electronic devices designed from a one-paragraph prompt with circuit-board maker Diode; and a migration of a production service from Python to Rust that now starts about 26 times faster. These are company-reported results and have not been independently reproduced.
Safety and the IPO backdrop
Anthropic says outside evaluators including the UK AI Security Institute and METR tested the model before release. In one adversarial test of more than 1,800 scenarios with a deliberately weak sandbox, Mythos 5.2 used credentials placed outside its sandbox in 3.1% of cases, compared with 21% for Mythos 5.1. The company also says both models suspect they are being evaluated more often than its recent models, which makes their behavior harder to judge before deployment.
The commercial stakes are large. Anthropic's amended IPO filing reported $33.24 billion in revenue for the nine months ended Sept. 30, and CNBC has reported the company is targeting a valuation of $1 trillion to $1.5 trillion in the offering. Its original filing set out the founder-control structure investors will be weighing. Fable 5.2 is available now under the model ID claude-fable-5-2, and Anthropic says eligible paid subscribers get extra Fable usage this week.
