AI News
All stories
Reported

Anthropic releases Claude Fable 5.2 and Mythos 5.2, cutting Fable API prices

Anthropic has released Claude Fable 5.2 and Mythos 5.2, which share one underlying model, and lowered Fable's API price to $8 and $40 per million tokens.

Models · · Updated · 3 min read

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, and cut the API price of Fable to $8 per million input tokens and $40 per million output tokens, down from $10 and $50. The two models share the same underlying model. What separates them is access. Anthropic says Fable has strong cybersecurity and biology safeguards that guard against misuse, which allows it to offer Fable to all users. Mythos is limited to vetted organizations in Anthropic's Life Sciences and Cyber Verification Programs, which the company recently expanded. Organizations pursuing beneficial work in cybersecurity or biomedical research can apply.

The safeguards determine what each model can be used for. Anthropic says Fable 5.2's cyber safeguards are as robust as those on Fable 5.1 but produce fewer unnecessary blocks. The company notes they will still block most cybersecurity-related tasks, which is why security work is directed toward Mythos. For most coding and knowledge work, Anthropic recommends starting with Claude Opus 5.5, its strongest daily driver. Users can switch to Fable or Mythos, or down to Claude Sonnet 5.5 or Claude Haiku 5.5, depending on the scope and complexity of the work.

FrontierCode measures whether a model's code changes would be merged by the maintainers of open-source repositories. At High effort, Fable 5.2 scores 45.3%. That is about level with Opus 5.5's best score of 45.5%, at roughly half the cost per task.

On computer-use and robotics tests, Fable 5.2 scores 89.3% on the OSWorld 2.1 offline set with partial credit. Fable 5.1 scored 83.5% and Opus 5.5 scored 87.2%. On RoboDojo Sim, Fable 5.2 reaches 64.9%, compared with 27.6% for Fable 5.1 and 39.4% for Opus 5.5. The table's footnote notes that RoboDojo Sim was run internally. On the Remote Labor Index, Fable 5.2 scores 41.3% at High effort, against 17.9% for Fable 5.1 and 21.3% for Opus 5.5.

Mythos 5.2 carries the security claims. It scores 89.6% on CVE Rediscovery, up from 78.3% for Mythos 5.1, and 76.3% on CyScenarioBench. CVE Rediscovery is Anthropic's own internal evaluation. It gives a model a snapshot of a project from before a published CVE was fixed, then measures how many published CVEs the model can independently find. Anthropic says its October 2026 run covered 1,829 CVEs. The company says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. In adversarial sandbox-escape tests, the model used planted credentials or files in 3.1% of scenarios, down from 21% for Mythos 5.1. The post also quotes a security company saying that one in six confirmed bugs found by Mythos 5.2 were missed by all of the other frontier models it tested.

Anthropic also described science and engineering results. Claude Fable 5.2 solved two open problems in theoretical physics concerning black holes. Claude Mythos 5.2, building on two open-source protein-folding models, found a way to predict the structure of whole adeno-associated virus capsids, accurately modeling 28 of the 34 it attempted. A team of 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in 30 hours. Anthropic's first formalization took 11 days. The post does not describe how the capsid or physics results were checked beyond the preprints it links.

Speed and cost changed alongside capability. Fable 5.2 generates output up to 30% faster than Fable 5.1, and Anthropic says testers found it communicates more clearly. Fast mode speeds up token generation by up to 2.5 times, priced at $16 per million input tokens and $80 per million output tokens. Medium is the default effort level in Claude apps and Claude Code. Anthropic says Fable 5.2 completes challenging tasks at Medium using fewer tokens than Fable 5.1 needed.

Most of the evidence is Anthropic's own. The benchmark table mixes conditions. Its footnote says Fable and Opus scores were run with safeguards on unless noted, and some figures are labeled without safeguards. The customer testimonials come from companies such as Shopify, Datadog and Uber. One Oracle test is more specific: two other models produced SQL that summed country-level amounts incorrectly, with values 5 to 8 times the expected figures, while Fable 5.2 returned the expected results.

What changed

The API price of Claude Fable 5.2 dropped to $8 per million input tokens and $40 per million output tokens, from $10 and $50, and Claude Mythos 5.2 is limited to vetted organizations.

Type: Pricing and limits · Takes effect: Monday, November 9, 2026 · Applies to: Claude Fable 5.2, Claude Mythos 5.2

Why it matters

Teams paying for Claude API calls on agentic coding or computer-use workloads get a lower per-token price. Mythos access, meanwhile, is limited to vetted organizations.

What you can do

Re-run cost estimates for current Claude workloads at the new Fable 5.2 prices and check whether Mythos eligibility applies to your organisation.

How the story developed

  1. Adds that the release came just ahead of Anthropic's IPO and that the company is back at the top of the AI race, per the report.

Original source

Further reporting

Where we saw it

2 sources

DAIR.AI (Elvis Saravia), THE DECODER

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works