Breaking

Anthropic launches Claude Fable 5.1 and restricted Mythos 5.1, cutting typical token costs about 25%

September 2, 2026 at 00:28 EDT

Foundation Models
Software Dev & Coding
AI Agents

Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, the same underlying model with different safeguards, positioning them as its most capable systems for coding, knowledge work, and long-running research. Fable 5.1 is generally available; Mythos 5.1 remains limited to trusted access programs for cybersecurity and life sciences.

Sep 1, 2026 · Anthropic

Fable 5.1 ships; Mythos 5.1 stays gated

Same underlying model, different safeguards. Cache reads drop 75% so typical token bills fall about 25%, and heavy agent jobs up to about 45%.

25%

typical cost cut vs Fable 5

45%

up to, on agentic cache-heavy work

$0.25

cache reads per million tokens

token context window

Terminal-Bench 4.0 (accuracy)

Fable 5.1 at 55.8% vs Fable 5 at 42.0%; Mythos 5.1 at 60.9%. Column height is proportional.

37.3%

GPT-5.6 Sol

42.0%

Fable 5

52.3%

Opus 5

55.8%

Fable 5.1

60.9%

Mythos 5.1

What testers liked

Faster, cheaper long-horizon coding; Cognition moving Devin traffic; rare-bug finds; more precise cyber/biology classifiers (about 60% fewer cyber false positives; 85% fewer benign biology blocks).

What still bites

Mythos stays U.S.-vetted. Fable still routes dual-use bio and some cyber (pentest, exploits, binary scanning) to Opus. EFS privacy is fall, not today. New accounts lose a distillation-friendly thinking-edit path.

Same model, two gates

1. Fable 5.1

GA on API and clouds

→

2. Safeguards

Cyber ID allowed; exploits blocked

→

3. Mythos 5.1

Trusted U.S. cyber and bio programs

Continue reading

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