Overview
On September 1, Anthropic unveiled Claude Fable 5.1 and Claude Mythos 5.1: two models that share one base model and differ only in safety guardrails. Fable 5.1 is generally available to all users, while Mythos 5.1 stays restricted to vetted organizations through the trusted access program. Anthropic calls them “the world’s most advanced models for coding and knowledge work.” Fable 5.1 scores 52.6% on the agentic science benchmark Terminal-Bench-Science 0.1, more than double Fable 5’s 24.7%, and cache-read prices drop 75%, cutting typical workload costs by about 25%.
One Model, Two Safety Layers
Like the Fable 5 / Mythos 5 pair from June, Fable 5.1 and Mythos 5.1 are the same underlying model behind different guardrails: Fable 5.1 routes high-risk requests in cybersecurity, biology and chemistry, and model distillation through classifiers and fallbacks, while Mythos 5.1 lifts some restrictions and is therefore offered only to vetted institutions. Both carry a 1M token context window and 128K max output tokens, with adaptive thinking always on. As OpenAI and Anthropic now consume about a third of new AI compute (our analysis), efficiency metrics on a release like this carry extra weight.
Benchmarks: Doubling Science-Agent Scores, Top Coding
- Agentic science: 52.6% on Terminal-Bench-Science 0.1, versus 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol;
- Coding: 55.8% on Terminal-Bench 4.0 (Fable 5: 42.0%), rising to 60.9% for Mythos 5.1 with looser guardrails;
- Knowledge work: 1853 on GDPval-AA v2, ahead of Opus 5 (1824) and Fable 5 (1723);
- Efficiency: matches Fable 5 accuracy with 20% fewer tokens on the internal finance benchmark at Anthropic, with big gains in slide generation. Open models are closing the gap through self-training (see how one model beat Opus 4.8 after two months of self-play, covered here), so frontier leads are thinning.
Pricing and API: 75% Cheaper Cache Reads, Three Breaking Changes
Base pricing is unchanged at $10 per million input tokens and $50 per million output tokens, but cache reads drop 75% to $0.25 per million tokens. Anthropic estimates total costs fall about 25% for typical workloads and up to 45% for highly agentic ones. The release also brings three breaking API changes that affect agents that edit conversation history; data retention defaults to 30 days, and zero retention requires express authorization.
Research Demos: Protein Binders, a Venus Elevation Map, GPU Kernels
- Protein design: Mythos 5.1 designed binders with a hit rate near 50% across 12 targets (industry typical: 10-15%), and roughly 10x higher binding affinity than the best Adaptyv Bio competition entry on three targets;
- Computational analysis: Fable 5.1 trained a neural network on NASA Magellan radar imagery to map a third of Venus at 2-3 km resolution instead of 10-20 km, with heights up to 25% more accurate;
- Computational biology: custom GPU kernels with cached intermediate results sped up seven open-source deep learning models by up to 2.5x, cutting expected GPU costs 30-60%. Like Evo generating living phage genomes for the first time (covered here), these demos point one way: models are moving from generating text to doing science directly.
FAQ
**What is the difference between Fable 5.1 and Mythos 5.1?**They share one base model and differ only in safety guardrails. Fable 5.1 is generally available (API model ID: claude-fable-5-1); Mythos 5.1 lifts some restrictions and is limited to vetted organizations.
**Did prices go up?**No. Input and output pricing is unchanged at $10/$50 per million tokens; cache reads are 75% cheaper at $0.25, so typical workloads cost about 25% less.
**Where can I use Fable 5.1?**It is available through the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Mythos 5.1 requires contacting an account team for access.
