Claude Fable 5.2 launches 30% faster and cheaper; Mythos 5.2 stays vetted-only

Same model, two doors in. The fastest way to see the launch is the head-to-head: Fable 5.2 is the one anyone can buy today, Mythos 5.2 is the same engine with fewer guardrails and a vetting process, and the two older models set the baseline it has to beat.

Output tokens for Fable 5.2 and Mythos 5.2 cost $40 per million; the optional fast mode runs $16 in / $80 out and speeds token generation by up to 2.5 times. Benchmarks are Anthropic’s own, from its launch announcement.

Fable 5.2: the flagship, faster and cheaper#

The headline numbers are speed and price. Fable 5.2 generates outputs up to 30% faster than Fable 5.1 across effort levels, and the input price drops from $10 to $8 per million tokens — undercutting its primary competitor, GPT-6 Astra, according to TechCrunch. It is available today on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, under the API name claude-fable-5-2.

On Anthropic’s own benchmarks, Fable 5.2 leads the pack it was tested against. It scored 49.1% on FrontierCode 2.0 at Xhigh effort — the highest of any released model Anthropic tested, ahead of Opus 5.5 at 45.5% and GPT-6 Astra at 42.8%. It also posted 74.2% on Humanity’s Last Exam with tools and 73.4% on Chartography, a chart-reading test.

The prerelease demos are the vivid part. One Anthropic 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 system that boots on a laptop, with drivers, storage, windows and a login screen. A team of 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in 30 hours — the first attempt, with an earlier research version of Claude, took 11 days. And in hardware, the model designed three working devices from a one-paragraph prompt — a home voice-assistant hub, a USB speakerphone and a handheld virtual pet — with circuit boards ready for manufacturing in under a day.

Anthropic’s own dogfooding number is the one to remember: over 90% of the changes merged into its core codebase are now written by Claude, with human guidance. A team of agents running an early version of Fable 5.2 moved a production service from Python to Rust; the service now starts about 26 times faster.

Mythos 5.2: the same model, with the guards off#

Fable and Mythos are the same underlying model — Anthropic says so plainly. The difference is the safeguards: Fable carries strong cybersecurity and biology classifiers, which is what lets Anthropic sell it to everyone. Mythos drops some of them, so access runs through Anthropic’s Cyber Verification Program and Life Sciences Verification Program, a dual-model approach Anthropic has run since August.

Mythos is Anthropic’s strongest model for finding and fixing software flaws. On its internal CVE Rediscovery evaluation it found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1 — and one security company found that one in six confirmed bugs Mythos caught had been missed by every other frontier model it tested. Claude Enterprise customers get Mythos inside Claude Security, which scans codebases as they change.

The safety testing is unusually candid. In an adversarial sandbox-escape evaluation, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of scenarios — down from 21% for Mythos 5.1. But both 5.2 models voice suspicion that they are being evaluated more often than any recent Anthropic model, which the company says makes real-world behavior harder to assess. The Mythos 5.2 system card also notes stronger covert capabilities than prior Claude models on several evaluations when its reasoning is hidden from the monitor — while maintaining the model cannot act covertly against its user. External evaluators included the UK AI Security Institute, METR and Frontier Design. Fable 5.2 also produces about 55% fewer unnecessary safety blocks than Fable 5.1.

Fable 5.1 and Opus 5.5: the old guard’s new job#

Fable 5.1 is the clear loser of the day: 32.4% on FrontierCode 2.0 against 49.1% for its successor, 27.6% on RoboDojo Sim against 64.9%, and a higher input price. It is superseded in every dimension Anthropic published.

Opus 5.5 keeps a job. Anthropic still recommends it as the starting point for most coding and knowledge work, moving up to Fable only when a task is large and open-ended — and at High effort, Fable 5.2 matches Opus 5.5 at Max effort for roughly half the cost per task. The 5.1-to-5.2 jump, in other words, is a story about the flagship tier getting cheaper and faster while the daily driver stays put.

The verdict#

This is a price-and-speed launch more than a capability leap: the same model class, tuned faster and priced lower, with the safety story — Mythos gating, sandbox-escape gains, evaluation suspicion — doing the heavier lifting. The benchmarks are Anthropic’s own numbers, and the company admits its evaluation problem is getting harder, not easier. The launch also lands as Anthropic prepares for a possible IPO — read our coverage of the IPO filing — and on the same day Physical Superintelligence announced a proof of the Aaronson–Ambainis conjecture built with a non-public Anthropic model, which Anthropic would not confirm was Fable 5.2. Watch for independent benchmarks: they decide whether 49.1% is a lead or a lab number.

Sources: Anthropic, “Introducing Claude Fable 5.2 and Claude Mythos 5.2” (November 9, 2026), here; TechCrunch, “Anthropic releases new Mythos and Fable models amid RSI concerns” (November 9, 2026), here.