Anthropic’s Claude Fable 5.1 and Mythos 5.1 share the same core model but differ in safeguard layers. Fable 5.1 is generally available via the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry, while Mythos 5.1 remains restricted to vetted US organizations in Project Glasswing. Both models support a 1 million‑token context window and 128 K maximum output tokens with adaptive thinking always enabled.
The headline performance shows a jump to 52.6 % on Terminal‑Bench‑Science 0.1, up from 24.7 % for the previous Fable 5 release and well above Opus 5 at 29.0 %. On Terminal‑Bench 4.0 Mythos 5.1 reaches 60.9 % versus 55.8 % for Fable 5.1, the gap reflecting fewer safeguard interventions rather than raw capability.
Cost savings come from a 75 % reduction in cache‑read pricing, dropping from $1.00 to $0.25 per million input tokens. Base input and output prices stay at $10 and $50 per million tokens respectively. For typical workloads this translates to roughly 25 % lower overall cost, and up to 45 % for agentic, context‑heavy use cases.
Teams migrating should note three breaking changes: forced tool use is removed (tool_choice = any or tool now returns 400), thinking blocks are model‑bound so earlier models cannot read Fable 5.1’s reasoning, and editing prior turns invalidates thinking blocks for accounts created after August 31 2026. Fixes include using auto tool choice with strict outputs, avoiding cross‑model fallback that relies on prior thinking, and employing turn‑scoped system messages or server‑side context edits when mid‑conversation changes are needed.
Safeguard updates allow vulnerability discovery but block exploit development, cutting cyber interventions by about 60 % per session. Biology safeguards trigger far less often on benign requests, while life‑science and cyber defence workloads still route to Opus when needed. Practical steps: enable cache‑read monitoring, adjust tool choice to auto, verify that turn‑level edits do not modify system or tools arrays, and test fallback paths to ensure reasoning is preserved when switching models.
#AI #Product #Claude #LLM #CostSaving #Safeguards