Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 today, three months after Fable 5 debuted to immediate export-control complications. The company describes them as the same underlying model shipped under different safeguard regimes: Fable 5.1 is generally available, while Mythos 5.1 remains restricted to vetted cybersecurity and life-sciences organizations through trusted access programs.
Benchmarks: The Science Numbers Stand Out
On Terminal-Bench-Science 0.1, a test of agentic scientific research tasks, Fable 5.1 scored 52.6% against Fable 5's 24.7%, Opus 5's 29.0%, and OpenAI's GPT-5.6 Sol's 22.4%. That is more than double its predecessor in under three months. On Terminal-Bench 4.0, a measure of terminal-based coding work, it hit 55.8% versus 42.0% for Fable 5. AutomationBench, designed around business workflows, nearly doubled as well: 31.4% versus 17.1%.
Anthropic is positioning the gains as early evidence that AI models will contribute directly to scientific progress. The company highlighted a test with investment firm Millennium, where Fable 5.1 reportedly traced a rare software crash to a bug in an external vendor library after the problem had stumped human engineers for four to five years.
Pricing: Cache Reads Down 75%
Headline token pricing is unchanged at $10 per million input tokens and $50 per million output tokens. The savings come from cache reads, which drop to $0.25 per million from Fable 5's $1. Anthropic estimates this cuts typical workload costs by roughly 25% and highly agentic workloads by up to 45%. Batch processing halves those rates again, to $5 and $25.
Fable 5.1 supports a one-million token context window with up to 128,000 tokens of output per response. Developers can call it using the model ID claude-fable-5-1 across the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry.
Enterprise Frontier Safeguards
Anthropic paired the model launch with Enterprise Frontier Safeguards, a new system that lets enterprise customers store activity data in their own cloud infrastructure rather than on Anthropic's systems. Misuse detection runs across sessions and accounts, routing alerts to the customer's security team rather than to Anthropic reviewers. The company developed EFS with more than 100 customers across financial services, healthcare, manufacturing, and the public sector.
Anthropic says it will not charge for EFS. The feature addresses friction that arose after the company introduced 30-day data retention with Fable 5. Customers in regulated industries found that policy difficult to accept. Rollout begins this fall; until then, eligible customers can run Fable 5 and Fable 5.1 under standard zero-data-retention terms.
Safeguards Tuned Down
Fable 5's safety classifiers triggered frequent complaints from users whose benign requests were being routed to fallback Opus models. Anthropic says the cybersecurity safeguards now flag about 60% fewer false positives, and biology and medical questions fall back around 85% less often. The model can now find software vulnerabilities, though it will not write exploits for them. Penetration testing and exploit generation still get redirected.
Developers can opt into automatic fallback via the Messages API, so users get a response from Opus 5 or Opus 4.8 instead of an error when a request is blocked.
Anti-Distillation Measures
Fable 5.1 ships with a lock on its reasoning. New API accounts created from today onward can no longer edit Claude's prior context in a multi-turn conversation while preserving the transcript of its thinking. Anthropic calls this a closing off of a publicly documented distillation technique. In February, the company traced over 16 million Claude exchanges to distillation campaigns run through roughly 24,000 fake accounts, naming three Chinese labs including Moonshot AI.
Existing accounts are not affected yet, but the restriction will apply to all users going forward with future model releases.
Watermarking Goes Global
Under the EU AI Act's Code of Practice on Transparency, Fable 5.1 and all models released after August 2, 2026 embed an imperceptible watermark in text outputs. The mark travels with copied text and may persist through light editing. Generated image files in PNG, JPG, and SVG formats carry signed provenance metadata following the C2PA standard. A detection API for the text watermark is in private preview for organizations eligible under EU law. Anthropic says neither feature affects output quality or content.
The company cautions that a detected mark indicates Claude processed the content but does not prove Claude authored it, since users often have Claude proofread, translate, or summarize human-written material.
What It Means
Fable 5.1 restores the Mythos tier's lead over Opus 5, which had narrowed the gap dramatically at half the price when it launched in July. Anthropic's documentation recommends Opus 5 for most workloads and Fable 5.1 for demanding reasoning and long-horizon agentic tasks where Opus falls short. The cache-read cut makes that recommendation easier to follow.
Fable 5.1 is available now to Pro, Max, Team, and Enterprise users on Claude, and to developers across all major cloud platforms.
