On September 1, Anthropic announced the release of Claude Fable 5.1 and Claude Mythos 5.1. The two models share the exact same foundation model but differ in safety guardrails: Fable 5.1 is open to all users, while Mythos 5.1 is available only through the Trusted Access Program to vetted cybersecurity and life sciences organizations.
Benchmark Surge: Scientific Agent Score Jumps from 24.7% to 52.6%
On the Terminal-Bench-Science 0.1 benchmark for agentic scientific research, Fable 5.1 scored 52.6% — more than double the previous generation Fable 5 at 24.7%, while Opus 5 scored 29.0% and GPT-5.6 Sol scored 22.4%.
On Terminal-Bench 4.0 for agentic coding, Fable 5.1 scored 55.8%, while Mythos 5.1 reached 60.9% under less restrictive safety settings.
In knowledge work, Fable 5.1 scored 1853 on GDPval-AA v2, surpassing Opus 5 (1824) and Fable 5 (1723).
On Artificial Analysis's latest LLM leaderboard, two variants of Fable 5.1 claimed the #1 and #2 spots, with Anthropic models occupying all top four positions.
More Affordable: Cache Read Costs Drop 75%, Up to 45% Savings on Agent-Heavy Workloads
Base pricing remains unchanged: $10 per million input tokens and $50 per million output tokens. However, cache read fees have been slashed by 75% to $0.25 per million tokens.
For typical workloads, Fable 5.1 is expected to cost about 25% less than Fable 5; for highly agentic workloads, savings can reach up to 45%.
Scientific Breakthroughs: From Protein Design to Venus Mapping
Anthropic demonstrated real scientific achievements using the new models:
Protein Design: Mythos 5.1, using open-source protein design tools, produced binders with 10x higher affinity than the best entries in a protein design competition across three targets; hit rate approached 50% across 12 targets, versus the industry typical 10-15%.
Computational Analysis: Fable 5.1 trained a neural network on radar data from NASA's Magellan mission from over 30 years ago, producing a high-resolution elevation map covering one-third of Venus — resolution refined from 10-20 km down to 2-3 km, with 25% improved accuracy.
Computational Biology: Mythos 5.1 rewrote GPU kernels for seven open-source deep learning models, accelerating inference by up to 2.5x, potentially reducing GPU costs by 30-60%.
Safety and Compliance: A Dual-Track Strategy
"Dual-Model" Approach: Fable 5.1 and Mythos 5.1 share identical weights but the latter features less restrictive safety guardrails, available only to vetted cybersecurity and life sciences organizations.
Fewer False Positives: The new cybersecurity guardrails block about 60% fewer false positives. Fable 5.1 can now be used to discover software vulnerabilities but cannot be used to develop exploits.
EU AI Act Compliance: Fable 5.1's outputs include invisible watermarks, meeting transparency requirements effective August 2, 2026.
Enterprise Frontier Safeguards (EFS): Combines zero-data-retention privacy with advanced security; customer data stored on infrastructure fully controlled by the enterprise, rolling out this fall.
The key highlight is the "same model, two safety levels" design — Anthropic is exploring a new product paradigm where the most powerful model isn't necessarily made available in the same form to everyone. With scientific agent capability doubling while long-horizon task costs drop 45%, Anthropic's optimization focus has shifted from "single-turn Q&A" to "continuously running agents." And with the IPO on the horizon, this release — blending performance breakthroughs, cost optimization, and safety grading — reads as a "model roadshow" aimed at the capital markets.