Breaking

Anthropic Releases Claude Fable 5.1, Cutting Cache Costs 75% and Raising Agentic Coding Scores

September 1, 2026 at 15:57 EDT

Foundation Models
Software Dev & Coding
AI Agents

Anthropic on Tuesday released Claude Fable 5.1, a generally available Mythos-class upgrade that the company says raises the bar on coding, knowledge work, and long-running problem solving while cutting the real-world cost of cached agent sessions. A twin model, Claude Mythos 5.1, shares the same capabilities with more permissive safeguards and remains limited to trusted access programs for cybersecurity and the life sciences.

September 1, 2026 · Anthropic

Claude Fable 5.1: cheaper cache, longer runs

Anthropic’s Mythos-class upgrade keeps $10 / $50 token rates, cuts cache reads 75%, and more than doubles agentic science scores versus Fable 5.

75%

Cheaper cache reads

~25%

Typical workload savings

52.6%

Terminal-Bench-Science

55.8%

Terminal-Bench 4.0

Terminal-Bench-Science 0.1

Stacked columns are proportional: 24.7% versus 52.6% — more than a 2× jump.

24.7%

Fable 5

52.6%

Fable 5.1

Cache-read price per million tokens

Four stacked blocks versus one: cache reads now cost one-quarter as much.

$1.00

Fable 5

$0.25

Fable 5.1

What testers liked

Cognition is moving Devin traffic from Opus 5 on launch day. Millennium said Fable 5.1 found a four-year crash no prior model explained.

Where it still lags

CodeRabbit saw review latency rise to 18:38 from 12:32. Vals.ai scored Harvey’s Legal Agent Benchmark at 6.67%, below Fable 5.

Same model, two doors

Fable 5.1

Generally available. Cyber false positives down ~60%. Benign biology fallbacks down ~85%.

Mythos 5.1

Trusted access only. Looser cyber and life-science safeguards for vetted defenders and scientists.

Continue reading

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