Anthropic launches Claude Fable 5.1 and Mythos 5.1 with up to 45% lower agentic costs
Anthropic launched Claude Fable 5.1 and Mythos 5.1 on Tuesday amid backlash over pricing, retention, and safeguards. Coverage places Fable 5.1 on the Claude API, AWS, Google Cloud, and Microsoft Foundry, with Mythos 5.1 limited to Project Glasswing vetted organizations. Anthropic cites 52.6% on Terminal-Bench-Science 0.1, doubling the prior score with over thirty percent better agentic coding. Pricing stays at $10 per million input and $50 per million output, while cache reads fall seventy-five percent to $0.25 per million; Anthropic estimates twenty-five percent savings on typical jobs and up to forty-five percent on agent-heavy runs. The release adds a statistical watermark and detection APIs for regulators and fact-checkers, influencing compliance planning. Those savings are unverified estimates, and The New Stack warns the watermark has a blind spot.
Anthropic launches Claude Fable 5.1 and Mythos 5.1 with up to 45% lower agentic costs
Anthropic launches Claude Fable 5.1 and Mythos 5.1, its most capable AI models yet. Fable 5.1 doubles its predecessor's score on Terminal-Bench-Science and improves agentic coding by over 30 percent.
Key takeaway
Anthropic's Fable 5.1 pairs higher agentic coding scores with up to 45% lower costs on tool-heavy runs, reshaping Claude's value for builders.
What happened
Anthropic launched Claude Fable 5.1 and Mythos 5.1 on Tuesday, calling them its most capable models yet while responding to customer criticism over price, data retention, and overzealous safeguards, according to The Verge.
Reporting cites 52.6% on Terminal-Bench-Science 0.1, a doubled predecessor score, over 30% better agentic coding, unchanged $10/$50 per-million token pricing, and a 75% cache-read cut to $0.25 per million tokens.
Evidence
Anthropic launched Fable 5.1 and Mythos 5.1 as its most capable models, with Fable 5.1 doubling Terminal-Bench-Science scores and improving agentic coding by over 30 percent.
The Decoder · attributed
Anthropic launches Claude Fable 5.1 and Mythos 5.1, its most capable AI models yet. Fable 5.1 doubles its predecessor's score on Terminal-Bench-Science and improves agentic coding by over 30 percent.
Fable 5.1 is estimated 25% cheaper than Fable 5 for typical workloads and up to 45% cheaper for highly agentic work.
Techmeme · attributed
Anthropic says Fable 5.1 will cost an estimated 25% less than Fable 5 for typical workloads and up to 45% less for highly agentic work
List token pricing is unchanged at $10/1M input and $50/1M output, with cache read pricing cut 75% to $0.25/1M tokens.
Techmeme · attributed
Anthropic prices Fable 5.1 at $10/1M input tokens and $50/1M output tokens, unchanged from Fable 5, and cuts cache read pricing by 75% to $0.25/1M tokens
Fable 5.1 is generally available on major cloud platforms while Mythos 5.1 stays restricted to vetted organizations under Project Glasswing.
MarkTechPost · attributed
Fable 5.1 is generally available on the Claude API, AWS, Google Cloud, and Microsoft Foundry; Mythos 5.1 remains restricted to vetted organizations under Project Glasswing.
Fable 5.1 ships with a statistical signature embedded in generated text and Anthropic is opening watermark detection to regulators and fact-checkers.
The New Stack AI · attributed
Anthropic launched Claude Fable 5.1 on Tuesday with a statistical signature embedded in its generated text
Anthropic is launching an API that lets regulators, media outlets, and researchers check whether text carries Claude's digital watermark.
The Decoder · attributed
Anthropic is launching an API that lets regulators, media outlets, and researchers check whether text carries Claude's digital watermark.
Why it matters
The release combines benchmark gains, aggressive pricing on cache-heavy agent loops, and watermark transparency tooling that will influence procurement, compliance planning, and competitive positioning in enterprise Claude deployments.
Limits and uncertainties
Cost savings figures of 25% to 45% are Anthropic estimates and have not been independently verified in the packet.
The New Stack reports the Fable 5.1 watermark has a blind spot developers should not ignore.
Critics cited in The Decoder warn watermarking could hurt text quality and create transparency problems where contracts ban AI use.
Practical implications
Teams running long autonomous agent workflows with heavy tool calls should rebenchmark total spend using the 75% cache-read price cut.
Builders in regulated or contract-sensitive environments must account for embedded watermarks and the new detection API in output and disclosure policies.
