Breaking

Anthropic launches Claude Fable 5.1 and cuts cached-token prices 75%

September 1, 2026 at 19:42 EDT

Foundation Models
AI Agents

Anthropic on Tuesday released Claude Fable 5.1, a generally available upgrade to its most capable model, alongside a restricted twin called Claude Mythos 5.1, and cut the price of cached context by 75% to lower the cost of long-running AI agents.

September 1, 2026 · Anthropic

Fable 5.1 makes long-running agents cheaper

Anthropic shipped Claude Fable 5.1 and restricted Mythos 5.1, cutting cache-read prices 75 percent while reducing false-positive safety blocks.

Cache reads now cost one-quarter as much

Each block equals $0.25 per million tokens. Shorter column, lower bill.

$1.00

Fable 5

$0.25

Fable 5.1

4x cheaper cache hits — the line item that dominates long agent loops.

75%

Cheaper cache reads

~25%

Typical workload cut

~45%

Highly agentic cut

60%

Fewer cyber blocks

Agentic science score more than doubles

Terminal-Bench-Science 0.1. Each block is about 5 percentage points.

24.7%

Fable 5

29.0%

Opus 5

52.6%

Fable 5.1

What testers like

Partners are shifting agent traffic onto Fable 5.1, citing cheaper cache hits, faster loops, and fewer false safety blocks on coding and research jobs.

What still constrains it

The $10 / $50 list price is still double Opus 5, Mythos stays gated, some sessions burn quotas quickly, and the new privacy controls do not ship until fall.

Enterprise Frontier Safeguards

Privacy equivalent to zero data retention, with misuse detection still in place. Rolling out later this fall; eligible customers get zero data retention until then.

Customer-owned cloud storage

→

Automated misuse detection

→

Customer reviews any flags

Continue reading

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