Breaking

Anthropic launches Claude Fable 5.1 with cheaper cache reads and mixed per-task costs

September 2, 2026 at 14:47 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic on September 1, 2026 released Claude Fable 5.1, a generally available Mythos-class model for long-horizon coding, research, and knowledge work, alongside a restricted twin, Claude Mythos 5.1. The pair share the same underlying weights; Fable 5.1 carries broader safety classifiers, while Mythos 5.1 is limited to vetted cybersecurity and life-sciences programs.

Sep 1, 2026 · Anthropic

Claude Fable 5.1 takes the frontier

Anthropic's Mythos-class update keeps $10/$50 list prices, cuts cache reads 75%, and leads independent intelligence scores — while using about 1.7x the output tokens of Fable 5 at max effort.

Intelligence Index at max effort

75%

Cache read price cut vs Fable 5

$10/$50

Input / output per million tokens

1.7×

Output tokens vs Fable 5 (max)

Terminal-Bench-Science 0.1

Accuracy, drawn to scale. Fable 5.1 more than doubles Fable 5.

52.6%

Fable 5.1

29.0%

Opus 5

24.7%

Fable 5

22.4%

GPT-5.6 Sol

Why cheaper cache is not cheaper tasks

At max effort, Fable 5.1 still costs about 20% more per Intelligence Index task because output tokens jump 1.7×.

1.0×

Fable 5 output

1.7×

Fable 5.1 output

Where it gains

Leads the Intelligence Index at 66
Science-bench accuracy more than doubles Fable 5
Fewer review comments, higher precision

Where it strains

About 20% more cost per max-effort task
Code review latency up 48.7%
Plan limits can run out mid-job

Live now as claude-fable-5-1

1M context · 128K max output · adaptive thinking always on · High default in Claude Code, Medium on Claude.ai. Mythos 5.1 stays invitation-only for cyber and life-sciences programs. Anthropic still points most workloads to Opus 5 first.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.