Summary: Anthropic shipped Fable 5.1 and Mythos 5.1 on September 1, 2026 as the same model with different safeguard levels. Fable 5.1 keeps list pricing but lowers cache costs and typical workload bills, while cyber false-positive interventions drop by about 60% per Claude Code session. Mythos 5.1 remains limited to trusted US organizations through cybersecurity and life-science verification programs.
On September 1, 2026, Anthropic shipped two models at once: Claude Fable 5.1 and Claude Mythos 5.1. What stands out about this release is the architecture rather than the scores — the company says outright that “Claude Fable 5.1 and Claude Mythos 5.1 are the same model, but with different levels of safeguards.” Fable 5.1 is generally available; Mythos 5.1 arrives only through trusted access programs, and its safeguards are built specifically for cybersecurity and life-science research.
One model shipped two ways — a pattern that mirrors how OpenAI would tier Astra through Daybreak a few days later: the frontier capability itself doesn’t move; what varies is who can reach it and under what guardrails.
Benchmark Numbers: Against Fable 5, Opus 5, and GPT-5.6 Sol
Anthropic describes Fable 5.1 as “the world’s most advanced model for coding and knowledge work.” Here are the comparison figures (Fable 5.1 / Fable 5 / Opus 5 / GPT-5.6 Sol):
- Terminal-Bench 4.0: 55.8% (Mythos 5.1: 60.9%) / 42.0% / 52.3% / 37.3%
- Terminal-Bench-Science 0.1: 52.6% / 24.7% / 29.0% / 22.4%
- Humanity’s Last Exam: 60.9% without tools, 65.0% with tools
- GDPval-AA v2: 1853 / 1723 / 1824 / 1711
- OSWorld 2.0: 77.9% partial, 41.7% strict
- AutomationBench: 31.4% / 17.1% / 26.9% / 19.6%
- CursorBench 3.2.0: 73.4% / 70.5% / 70.0% / 67.2%
On Terminal-Bench, Mythos lands 5 points ahead of Fable. Anthropic’s read: that gap reflects where the older, less precise cyber safeguards kicked in, and the update is expected to narrow it.
Pricing: List Prices Hold, Bills Shrink
Input remains $10 per million tokens and output $50, but two adjustments pull real costs down: cache reads fall to $0.25 per million (75% cheaper), typical workloads come in about 25% cheaper than Fable 5, and heavily agentic workloads save up to roughly 45%. Effort defaults to High in Claude Code, Medium in Claude Cowork and Claude.ai. In the words of Every CEO Dan Shipper: “It’s friendly Fable. Fable-level intelligence, Opus-level price, Sonnet-speed.”
Guardrail Tuning: Fewer False Positives, Not a Blanket Loosening
Precision is the product-relevant change here. On the cyber side, false-positive interventions fall by roughly 60% per Claude Code session; Fable 5.1 “can now be used to discover software vulnerabilities — though not to develop exploits for them,” while dual-use work like penetration testing still routes to Opus models. On the biology side, safeguards trigger 85% less often on benign elementary biology and medical questions, though life-science R&D still routes to Opus unless you come through Mythos.
The safety disclosures deserve recording too: Mythos 5.1 shows greater bio capability than Mythos 5 while remaining below the next Responsible Scaling Policy risk tier, and its cyber capability — the strongest Anthropic has shipped — stays inside a lower Frontier Compliance Framework risk category. A new anti-distillation measure prevents new API accounts from editing prior context while preserving Claude’s thinking transcript.
Mythos 5.1 and Trusted Access: Who Gets It, and For What
Today Mythos 5.1 is restricted to a set of US organizations, with wider availability being coordinated with the US government. There are two channels: the Cyber Verification Program (CVP, reduced cyber safeguards for defensive security work) and the Life Sciences Verification Program (LSVP, developed in partnership with the US government, first participants already enrolled). Claude Security’s codebase vulnerability scanning now runs on Mythos 5.1. Enterprises additionally get Enterprise Frontier Safeguards (EFS), which keeps data on customer-controlled cloud infrastructure, rolling out in phases starting fall 2026.
Three research cases show what trusted access yields: on three Adaptyv Bio protein targets (EGFR, Nipah G, 15-PGDH), Mythos 5.1 reached binding affinities 10x higher than the best competition entries, with a near-50% hit rate across 12 targets versus a typical 10–15%; Fable 5.1 used Magellan radar data to rebuild a Venus elevation map, sharpening resolution from 10–20 km to 2–3 km and releasing it under Creative Commons; and Mythos 5.1 accelerated seven open-source genomics and protein models up to 2.5x on an H100, an estimated 30–60% cut in genome-wide GPU costs.