- Sources: primary, discussion
- Summary: Anthropic published the release on 2026-09-01 and states Fable 5.1 costs an estimated 25 percent less than Fable 5 for typical token-billed workloads and up to approximately 45 percent less for highly agentic work, entirely through reduced cache-read pricing. Anthropic states Fable 5.1 can be used to discover software vulnerabilities but not to develop exploits for them, that its newest cybersecurity safeguards block 60 percent fewer false positives, and that Mythos 5.1 is the same underlying model with different safeguards, available only through trusted access programs for cybersecurity and life sciences work. Anthropic reports Terminal-Bench 4.0 at 55.8 percent for Fable 5.1 and 60.9 percent for Mythos 5.1 against 42.0 percent for Opus 5, states the gap between the two reflects tasks where its earlier cyber safeguards intervened, and states that Fable 5.1 was evaluated with production safeguards enabled, that Fable 5.1 and Fable 5 scored zero on OSWorld 2.0 on tasks where those safeguards intervened, and that in all other safeguard interventions the cybersecurity tasks were completed by Claude Opus 4.8 and the biology tasks by Claude Opus 5.
- Why it matters: The cache-read price change moves the unit economics of long agent runs, and the safeguard change moves what security work the model will do at all.
- Follow-up: Track whether independent benchmarks reproduce the Terminal-Bench 4.0 figures Anthropic reports.
send feedback on this story