AI Change Tracker
4

Anthropic launches Claude Fable 5.1 and Mythos 5.1 with cache price cut, removed data retention limits

2026-09-01

On September 1, 2026, Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, positioned for long-running agentic coding, knowledge work, and research, both with a 1M-token context window, 128k max output tokens, and always-on adaptive thinking. List pricing stays at $10/$50 per MTok input/output (same as Fable 5), but prompt cache read price drops 75% to $0.25/MTok. Anthropic also introduced Enterprise Frontier Safeguards (EFS) for agent observability. Per Stratechery, Fable's prior data retention policy was removed rather than merely altered. Anthropic's own benchmark table reports Fable 5.1 scoring 52.6% on the new Terminal-Bench-Science 0.1 benchmark versus 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol. TechCrunch reported the release also reduces false-positive safeguard restrictions. Per Latent Space's AINews recap, Artificial Analysis measured roughly 1.7x more output tokens per task, offsetting the cache savings for a reported ~20% net per-task cost increase.

Significance 4: Flagship release from a tracked lab with real pricing changes and lab-reported benchmark claims beating current flagship comparisons; capped at 4 pending independent reproduction of the new benchmark.

technology financial business Release Pricing change Lab strategy

Implications · machine-drafted, not owner judgment

If Fable 5.1's coding/agentic gains hold up independently, it becomes a live challenger to Opus 4.8 for the coding-agent and daily-driver picks; the 75%-cheaper cache reads make long, cached agent sessions materially cheaper even though the observed ~20% net per-task cost increase (from higher output-token usage) partly offsets that on shorter tasks. Removing the data-retention limits and adding Enterprise Frontier Safeguards targets exactly the zero-data-retention objection that has slowed some enterprise Claude adoption, which could accelerate enterprise deals. For the owner, this is a candidate axis update once independent SWE-bench/Terminal-Bench numbers land, and a case study in a lab responding to an enterprise trust objection with policy rather than announcement alone.

Watch for
  • Independent SWE-bench Verified / Terminal-Bench run on Fable 5.1 within a month
  • Independent measurement of real agentic-workload cost (not just cache-read list price)
  • Enterprise adoption commentary citing the removed data-retention limit as a deciding factor

Sources

Models

Affected axes

Coding · Agentic / tool use · Hard reasoning · Cost

← Back to feed