Anthropic ships Claude Fable 5.1 and Mythos 5.1 with signed chain-of-thought security
The AMW Read
Anthropic, an existing top-tier case-study lab, updates its frontier model line with benchmark claims, a steep agentic-cost cut, and a genuinely new signed-reasoning anti-tampering mechanism, but introduces no new entrant or debate resolution.
Anthropic ships Claude Fable 5.1 and Mythos 5.1 with signed chain-of-thought security
Anthropic released Claude Fable 5.1 and a restricted-access Mythos 5.1, claiming top results across eight public benchmarks with the largest gains in scientific research and coding versus both Fable 5 and GPT-5.6 Sol. Pricing changes center on cached-token reads, cut 75% to $0.25 per million tokens, while input ($10/M) and output ($50/M) prices hold steady; Anthropic says typical workloads run about 25% cheaper than Fable 5, rising to 45% for heavily agentic tasks where cache reads dominate cost. Fable 5.1 is generally available; Mythos 5.1 is limited to vetted cybersecurity and life-sciences organizations through a trusted-access program, with vulnerability discovery now permitted (exploit development still routed to Opus).
The more consequential change is a new anti-distillation control: each chain-of-thought block now carries a cryptographic signature verifying both model authorization and that the preceding conversation, system prompt, and tool list are unaltered, closing a documented technique for editing a model's prior context while preserving its reasoning trace to replay it under adversarial instructions. New API accounts get this enforced by default; Claude Code, claude.ai, and the Agent SDK already comply automatically. Anthropic is tracked in the AI Market Watch index with $132.3 billion in total funding (coverage, not a census), underscoring the capital riding on frontier labs differentiating on cost and trust controls rather than raw scale alone.
Developers calling the Messages API directly need to treat conversation history as append-only and test the `prefix_mismatch_behavior` parameter before enforcement extends to existing keys, which are currently grandfathered if registered before August 31. For investors, the steep cache-price cut specifically rewards long-context agentic workloads, reinforcing that near-term competitive pressure among frontier labs is concentrated on agentic price-performance and API-level trust guarantees rather than headline parameter counts.
#Anthropic #ClaudeFable5.1 #FoundationModels #AIAgents #ModelSecurity #APIpricing
