Claude Fable 5.1: Benchmark Gains and 25-45% Lower Cost

2 min read
anthropicmodel-releasellm-benchmarkspricing
View as Markdown
Originally from tiktok.com
View source

My notes

Watch on TikTok Tap to open video

Summary

A screen-share walkthrough of Anthropic’s announcement post for Claude Fable 5.1, with a parallel more-permissive Claude Mythos 5.1 tier, covering benchmark gains over Fable 5 and rival GPT-5.6 Sol, a 25-45% cost cut via cheaper cache reads, a new customer-controlled-infrastructure privacy option, and safety changes including anti-distillation protections. The source is a creator scrolling and narrating the actual post rather than adding independent commentary.

Key Insight

What Fable and Mythos are

Per the post, Claude Fable 5.1 and Claude Mythos 5.1 are the same underlying model. Mythos is a trusted-access variant with more permissive safeguards for vetted individuals and organizations; Fable is the standard-safeguard public release.

Benchmarks (Fable 5.1 / Fable 5 / Opus 5 / GPT-5.6 Sol)

  • Terminal-Bench-Science 0.1: 52.6% / 24.7% / 29.0% / 22.4%
  • Terminal-Bench 4.0 (agentic coding): 55.8% / 42.0% / 52.3% / 37.3%
  • GDPval-AA v2 (knowledge work, Elo-style): 1853 / 1723 / 1824 / 1711
  • OSWorld 2.0 (computer use, partial): 77.9% / 72.9% / 75.9% / not reported
  • Humanity’s Last Exam (no tools): 60.9% / 57.8% / 56.6% / not reported
  • AutomationBench: 31.4% / 17.1% / 26.9% / 19.6%
  • CursorBench 3.2.0: 73.4% / 70.5% / 70.0% / 67.2%

Fable 5.1 leads Opus 5, Anthropic’s own higher tier, on every listed benchmark, and leads GPT-5.6 Sol everywhere it’s reported.

Pricing

Cache reads now cost $0.25 per million tokens, a 75% cut on that line item. Net effect is roughly 25% cheaper for typical workloads and up to 45% cheaper for complex coding and highly agentic tasks, because agentic workloads lean more heavily on cache reads. The cost index shows Fable 5 at 100 versus Fable 5.1 at 75 for typical workloads and 55 for highly agentic ones, measured over four weeks of actual August 2026 usage.

Data retention and privacy

A new Enterprise Frontier Safeguards system stores enterprise customer data in infrastructure the customer controls rather than Anthropic’s, giving zero-data-retention-equivalent privacy while keeping adversarial-use protections. It rolls out in phases starting later in fall 2026; until then, eligible customers can already run Fable 5.1 with standard zero data retention.

Safeguards changes

  • Cybersecurity false-positive rate down roughly 60% versus Fable 5.
  • Fable 5.1 can now be used to discover software vulnerabilities for defensive security work but is still blocked from developing exploits. Dual-use tasks still route to Opus models.
  • New anti-distillation protection: for API accounts created from the 5.1 launch date onward, it’s no longer possible to manually edit Claude’s prior turns in a multi-turn conversation while preserving the transcript of extended thinking. This closes a publicly known technique for distilling Claude’s reasoning via edited conversation history. Existing accounts are unaffected for now.
  • A new biology access program built with the US government will enable Mythos 5.1’s advanced biology capabilities for vetted scientists.

Availability

Live on AWS, Google Cloud, and Microsoft Azure alongside the Claude API directly, with the API model identifier claude-fable-5-1.