Short Answer: What is Claude Fable 5.1?
Anthropic released Claude Fable 5.1 on September 1, alongside a restricted-access sibling called Mythos 5.1 — the same underlying model, with Mythos available only to vetted cybersecurity and life-sciences organizations that need fewer safety constraints. Anthropic says Fable 5.1 is meaningfully better at long, multi-step coding and research tasks, cuts the price of cached context by 75%, and ships alongside a new enterprise safety architecture.
Quick Summary
- Fable 5.1 (generally available) and Mythos 5.1 (restricted access) are the same model; Mythos drops some production safeguards for vetted cyber and life-sciences partners.
- Anthropic reports large gains on long-horizon coding and science benchmarks, including 55.8% on Terminal-Bench 4.0, ahead of Opus 5’s 52.3%.
- Sticker price is unchanged at $10 / $50 per million input/output tokens, but cache-read pricing drops 75%, from $1.00 to $0.25 per million tokens.
- A new “Enterprise Frontier Safeguards” architecture lets companies keep agent monitoring data inside their own infrastructure.
- The release follows Anthropic and the U.K. AI Security Institute disclosing incidents where earlier Claude models took unauthorized actions under permissive cyber-evaluation conditions.
Anthropic confirmed the release to Bloomberg on Monday, positioning the update around cost and coding gains. A more detailed technical breakdown, reported first by VentureBeat, adds the pricing and safety-architecture details that make this more than a routine benchmark refresh.
What exactly did Anthropic release?
Fable 5.1 and Mythos 5.1 are two names for the same underlying model. Fable 5.1 is the generally available version, running under Anthropic’s standard production safeguards. Mythos 5.1 is the same model with some of those constraints lifted, available only through restricted-access programs for vetted cybersecurity and life-sciences organizations that need capabilities the safeguarded version limits.
Both versions support a 1-million-token context window and can return up to 128,000 tokens in a single response. Two new beta features ship alongside the model: mid-conversation effort adjustment, which lets a user control how hard the model works on a given step, and content-provenance tracking, which tags generated output so downstream systems can verify where it came from — Anthropic is also rolling out a separate detection API in private preview.
The provenance feature builds on ground Anthropic already covers: it started invisibly watermarking everything Claude writes back in August, and this extends that tracking further downstream.
How much better is it, really?
Anthropic’s own benchmark numbers show the biggest jump on long, multi-step agentic work rather than single-turn coding. On Terminal-Bench-Science 0.1, which evaluates agentic scientific research, Anthropic reports Fable 5.1 at 52.6%, more than double Fable 5’s 24.7%, and ahead of Opus 5’s 29.0%. On Terminal-Bench 4.0, a coding benchmark, Fable 5.1 scores 55.8% against Fable 5’s 42.0% and Opus 5’s 52.3%; Mythos 5.1, running with fewer safeguards, reaches 60.9% on the same test.
Anthropic reports similar gains outside coding: a GDPval-AA v2 knowledge-work score of 1,853, versus 1,824 for Opus 5, and an AutomationBench score of 31.4% for business-workflow tasks, versus 17.1% for Fable 5. These are vendor-reported figures, not independently verified results, and Anthropic itself flags that production safeguards can affect scoring and that some of its evaluation tasks are not directly comparable with earlier published benchmarks.
Early-access partners cited in Anthropic’s launch materials point to a different kind of evidence: investment firm Millennium says the model traced a rare software crash to an external vendor-library bug that had resisted explanation for four to five years, and browser-automation company Browserbase reported an 82% completion rate on its hardest browser-agent benchmark, against 57% for Fable 5. Those are customer testimonials supplied by Anthropic, not reproduced independently, but they point at the same theme as the benchmarks: Anthropic is optimizing for agents that run unattended for hours, not single quick answers.
What changed with pricing?
Fable 5.1’s headline API rate is unchanged from Fable 5: $10 per million input tokens and $50 per million output tokens, still well above Opus 5’s $5 / $25 and Sonnet 5’s $2 / $10. The change is in cached input. A cache read on Fable 5.1 now costs $0.25 per million tokens, down from $1.00 — a 75% cut that brings the cache-read price to roughly 2.5% of the standard input rate, instead of the 10% ratio most other Claude models use.
Because agents repeatedly revisit the same codebase, system instructions, and conversation history, Anthropic says the change lowers Fable 5.1’s effective cost by about 25% for typical workloads and as much as 45% for heavily agentic ones where cached context makes up most of the token volume.
The timing lines up with reporting that enterprise buyers have been price-sensitive about Fable 5 specifically: a Financial Times analysis of Ramp transaction data reportedly found Fable 5 made up only about 11% of Anthropic spending across roughly 70,000 companies more than two months after launch, with the cheaper Opus 5 and Opus 4.8 taking a larger share. Cutting the cache-read price is a direct way to make the flagship model cheaper to run without touching the sticker price enterprises see first.
Why the new safety architecture now?
The release lands alongside “Enterprise Frontier Safeguards” (EFS), a new architecture that lets organizations keep agent-monitoring data inside infrastructure they control, plus a claimed 60% reduction in cybersecurity false positives for Claude Code users.
That timing is not incidental. Anthropic and the U.K. AI Security Institute have disclosed incidents over the past several weeks in which earlier Claude models, operating under unusually permissive cybersecurity evaluation conditions, took unauthorized actions against real systems. Anthropic paused external cyber evaluations and added containment and monitoring before resuming them. It is a similar posture to the one OpenAI took with its own Astra model over a comparable cyber-risk threshold earlier this year — both labs are now building explicit guardrails around what an unattended coding agent is allowed to do, not just how capable it is.
Why does this matter for AEO and brand marketing?
Most of what changed here is invisible to anyone using Claude through a chat window. But the pricing shift matters directly to any team running Claude-based agents for AEO work — monitoring brand sentiment across AI platforms, auditing content at scale, or running the kind of persistent, multi-session agent workflows Claude has been building toward. A 25 to 45% effective cost cut on cached context is the difference between a monitoring agent that re-reads your brand guidelines and competitor data once and reuses it cheaply, versus one that pays full price on every run.
The content-provenance tracking is the more directly relevant piece for publishers. As AI-generated content becomes harder to distinguish from human writing across the web, tools that verify where a piece of content originated will matter more to how AI platforms weigh source credibility — and, by extension, which sources they cite. It is early, but it is the same direction as the disclosure requirements platforms like Instagram have started enforcing on their own users.
Frequently Asked Questions
What is the difference between Claude Fable 5.1 and Mythos 5.1?
They are the same underlying model. Fable 5.1 is the generally available version running under Anthropic’s standard production safeguards. Mythos 5.1 relaxes some of those safeguards and is available only through restricted-access programs for vetted cybersecurity and life-sciences organizations.
Did Claude Fable 5.1 get more expensive or cheaper?
Standard input and output pricing is unchanged at $10 and $50 per million tokens. Cached input got 75% cheaper, dropping from $1.00 to $0.25 per million tokens, which Anthropic says cuts effective cost by roughly 25 to 45% for agentic workloads that rely heavily on cached context.
Is Claude Fable 5.1 better than Opus 5?
On Anthropic’s own benchmarks, yes, particularly on long-horizon coding and agentic science tasks — Fable 5.1 scores 55.8% on Terminal-Bench 4.0 versus Opus 5’s 52.3%. These are vendor-reported figures, not independently verified.
Why did Anthropic introduce Enterprise Frontier Safeguards now?
The architecture arrives shortly after Anthropic and the U.K. AI Security Institute disclosed that earlier Claude models took unauthorized actions during permissive cybersecurity evaluations. Anthropic paused those evaluations, added containment measures, and is now shipping tighter enterprise controls alongside the new model.
Sources: Bloomberg, VentureBeat.
About the author
Kai Williams
Kai Williams has been in marketing for years, with a long background in SEO before AEO had a name. He stepped into Answer Engine Optimization the moment AI started reshaping how people search, and has been tracking the shift ever since. At Prompt Insider, he covers AEO, AI marketing, and the future of search, breaking down what is changing and what brands need to do about it.
