Anthropic Releases Claude Fable 5.1 and Mythos 5.1
Anthropic launches Claude Fable 5.1 and Mythos 5.1, hitting 52.6% on Terminal-Bench-Science and cutting cache reads by 75%.
Stock photo for illustration only, not from the actual event
- Claude Fable 5.1 is now generally available across multiple cloud platforms.
- Claude Mythos 5.1 is restricted to vetted US organizations under Project Glasswing.
- Fable 5.1 scores 52.6% on Terminal-Bench-Science 0.1.
- Cache read pricing drops 75% down to $0.25 per million tokens.
Anthropic has officially released its newest artificial intelligence models, Claude Fable 5.1 and Claude Mythos 5.1. Claude Fable 5.1 is generally available as claude-fable-5-1 across the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. Meanwhile, Claude Mythos 5.1 is restricted exclusively to vetted US organizations participating in Project Glasswing.
On Terminal-Bench-Science 0.1, an agentic scientific research benchmark, Fable 5.1 achieved a score of 52.6%, outperforming Opus 5 at 29.0%, Fable 5 at 24.7%, and GPT-5.6 Sol at 22.4%. Anthropic reported a standard error ranging from 3.5 to 4.5 points per model, advising users to interpret the margin rather than the exact ranking with care.
Stock photo for illustration only, not from the actual event
On Terminal-Bench 4.0, Fable 5.1 reached 55.8% and Mythos 5.1 reached 60.9%. The performance gap between two identical models represents the cost of safeguard interventions, serving as an unusually honest disclosure. Additional benchmark results include CursorBench 3.2.0 at 73.4%, Humanity’s Last Exam at 60.9% without tools and 65.0% with tools, AutomationBench at 31.4%, OSWorld 2.0 strict at 41.7%, and GDPval-AA v2 at 1853.
Base input and output pricing remain unchanged, but cache read costs dropped significantly by 75%, falling from $1.00 to $0.25 per million tokens. This equals 0.025 times the base input cost compared to 0.1 on all other Claude models. Anthropic measures roughly 25% lower costs on typical workloads and up to about 45% on context-heavy agentic tasks, while batch processing is priced at $5 and $25 per million tokens.
Anthropic's massive 75% reduction in cache read pricing addresses a major economic hurdle for developers running complex, context-heavy AI agents. By significantly lowering the cost of reusing cached prompt data, developers can scale multi-turn autonomous workflows much more economically.
Additive changes in beta behind headers include per-message effort, turn-scoped system messages, and thinking.display: "updates". Content provenance is enforced rather than optional, applying a statistical text watermark to all outputs and C2PA credentials on files.
In research applications, Mythos 5.1 designed protein binders with an approximate 50% hit rate across 12 targets against a standard 10% to 15% norm. Fable 5.1 built a Venus elevation map at a 2 to 3 km resolution, and custom GPU kernels accelerated seven open-source genomics models by up to 2.5x.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment