SHELFCRITTER AI Digest

MODEL

Claude Mythos 5.1

modeltopic-noteanthropicsecurity

Overview

Claude Mythos 5.1 is Anthropic’s 2026-09-01 mid-cycle refresh of Claude Mythos 5, shipped alongside its sibling Claude Fable 5.1 as the same underlying model with different guardrails — Mythos 5.1 is the restricted-access variant for vetted cyber and life-sciences organizations while Fable 5.1 is the GA release with production safeguards. The load-bearing change on the 5.1 cycle is cache economics — cache-read drops to $0.25/MTok, a 75% cut and now 2.5% of input price — applied to the same token pricing ($10/$50 per MTok) and same 1M context / 128K output as the Fable sibling. Anthropic-disclosed benchmarks: Mythos 5.1 posts 60.9% on Terminal-Bench 4.0. Access continues under the Glasswing / NSA distribution shape Claude Mythos 5 established.

Timeline

  • 2026-09-02-AI-Digest — Anthropic releases Claude Mythos 5.1 on 2026-09-01 alongside Claude Fable 5.1 — same underlying model, differing only in the guardrail layer. Anthropic-disclosed benchmark: Mythos 5.1 posts 60.9% on Terminal-Bench 4.0. Cache-read economics land the same 75% cut to $0.25/MTok as the Fable sibling; token pricing held at $10/$50 per MTok on 1M context / 128K output. HN reception 1,057 pts / 985 cmts on the joint release. The digest’s disciplined framing to carry: the 5.1 refresh is the second instance of the sibling-SKU deployment pattern Claude Mythos 5 / Claude Fable 5 introduced in June 2026 — Mythos remains the restricted-access variant for vetted cyber and life-sciences organizations while Fable ships as GA with production safeguards. Log against MOC - Agent Security and MOC - Major Companies.

Key Developments

  1. Same Underlying Model as Fable 5.1, Guardrail-Split Not Weight-Split (September 1, 2026): Mythos 5.1 and Claude Fable 5.1 are the same underlying model with different guardrails — the deployment mechanism is the guardrail layer, not a capability tier. Extends the Claude Fable 5 / Claude Mythos 5 classifier-mediated deployment primitive into the 5.1 refresh at the same $10/$50 per MTok pricing and 1M context / 128K output shape.

  2. 60.9% on Terminal-Bench 4.0 as the Disclosed Capability Datum: Anthropic-disclosed benchmark result; log as the anchor number the Mythos-tier corpus should cite for the 5.1 refresh. Terminal-Bench 4.0 sibling number for Fable 5.1 is 55.8%; the Mythos vs Fable delta on the same eval is small — consistent with the shared-model framing where the SKU split is guardrail, not capability.

  3. Cache-Read $0.25/MTok / 75% Cut Applies Identically Across the Sibling Pair: The pricing-substrate change lands on both SKUs symmetrically — the 5.1 cycle’s load-bearing story is the cache economics, and both siblings ride the same 75% cache-read cut to $0.25/MTok (2.5% of input price). The distribution asymmetry (Glasswing / NSA gated for Mythos vs GA for Fable) is unchanged from the 5.0 cycle.

See also: Anthropic, Claude Fable 5.1, Claude Mythos 5, Claude Fable 5, Project Glasswing, Claude Mythos Preview, MOC - Major Companies, MOC - Agent Security.