Claude Fable 5.1 and Mythos 5.1: What Changed, Who Gets Access, and What It Costs

What launched on September 1, 2026

The pairing follows a pattern Anthropic set earlier in the year. Mythos first appeared as Mythos Preview on April 7, 2026, alongside Project Glasswing, a program that gave a small group of organizations, including AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, access to the model to find and fix vulnerabilities in critical software. Fable 5 then became the broadly available version. With the 5.1 release, both arrive on the same day.

What is new in Fable 5.1 compared with Fable 5

• Longer unattended work. Anthropic highlights sustained problem-solving over many hours, including a 38-hour machine-learning run in which the model designed and ran its own experiments.
• Cheaper cache reads. Reading cached context now costs $0.25 per million tokens, down 75% from $1.00 on Fable 5. Cache reads on Fable 5.1 cost 2.5% of the input price, versus 10% on most Claude models, which matters for long Cowork sessions that reread the same files.
• Fewer false alarms on security work. Anthropic reports about 60% fewer cyber safeguard interventions per Claude Code session than with Fable 5. Fable 5.1 can help find vulnerabilities, but it still hands exploit development off rather than writing it.
• Same list price. Input and output stay at $10 and $50 per million tokens.

The AutomationBench jump is the most relevant number for Cowork users, because it measures the kind of multi-step business task Cowork exists for: pulling data from several systems, updating records, and producing a finished output. Fable 5.1 nearly doubles Fable 5's score.

Fable 5.1 on Claude plans: included or usage credits

Fable 5.1 is billed the same way Fable 5 has been since July 20, 2026, and it works differently from every other Claude model:

Two details trip people up. First, the 50% on Max is a ceiling inside your existing allowance, not a second pool on top of it. Second, Fable uses that allowance faster than other models because each task consumes more. A long Cowork task on Fable can use a large share of a five-hour or weekly limit in one go.

A quick way to estimate what a task costs

Look at how much a task reads and how much it writes. A task that reads 200,000 tokens of documents and writes a 10,000-token report costs about $2 in input and $0.50 in output at API rates, before caching. Rerunning it against the same cached files costs far less, because cached reads are billed at $0.25 per million tokens. Agentic tasks often loop, rereading files and calling tools many times, so a real session can cost several times this simple estimate. Run one representative task, note what it consumed, and multiply by how often you plan to run it.

Fable 5.1 vs Opus 5.5: is Fable worth it?

Three weeks after Fable 5.1 shipped, Anthropic released Opus 5.5 and said it performs at Fable 5.1's level on most work at $4 input and $20 output per million tokens, 40% of Fable's price. On Terminal-Bench 4.0, Opus 5.5 even scores higher: 66.4% against Fable 5.1's 55.8%. For most Cowork tasks, Opus 5.5 is now the sensible default.

This matches Anthropic's own developer guidance: start with Opus 5.5, and move up to Fable 5.1 for demanding reasoning and long-horizon agent work, or when Opus 5.5 at higher effort still falls short. Switch models per conversation in the model picker, which stayed in the same place after chat and Cowork merged.

Mythos 5.1: who can use it and why it is restricted

Mythos 5.1 is the same model as Fable 5.1 with safeguards relaxed for two kinds of work where the standard rules get in the way of legitimate experts: cybersecurity and the life sciences. Anthropic keeps it behind verification because the same abilities that help defenders and researchers could help attackers.

Why Mythos is gated

Mythos models have shown unusually strong security skills. During Project Glasswing, partners used Mythos Preview to scan critical codebases for vulnerabilities. In July 2026, Anthropic reported that Mythos Preview found previously unknown mathematical weaknesses in cryptographic algorithms, including HAWK, a candidate signature scheme for the post-quantum era, and a faster method for testing AES. Findings like these are why Anthropic offers Mythos only to vetted organizations.

The verification programs

• Cyber Verification Program: for security teams and defenders who need help with vulnerability research that standard safeguards would block.
• Life Sciences Verification Program (announced September 17, 2026): organizations pass a review of research credentials, security standards, and ethical oversight. A Standard Use Grant unlocks Mythos 5.1, Opus 5, and Sonnet 5 with classifiers that are more permissive for science tasks and renews yearly. A High-Risk Use Grant removes the life-sciences safeguards for a single project and must be renewed every six months; Mythos is excluded from that tier. Cyber classifiers stay on under every grant.

Access is currently limited to a set of US organizations, and the life-sciences program is open to Claude Enterprise, Team, and API customers, not individual Pro or Max subscribers. It is also not yet available through third-party cloud platforms.

Notes for enterprise teams

• Enterprise Frontier Safeguards. Anthropic is rolling out a way to keep safeguard monitoring data inside your own AWS, Azure, or Google Cloud environment, under your encryption and access controls, starting in fall 2026.
• Budget for Fable separately. Because Fable draws allowance or credits faster, decide which teams get it and for which work, rather than letting it become everyone's default.
• Retirement date. Anthropic commits to keeping Fable 5.1 available until at least September 1, 2027, so workflows built on it are not at risk soon.

If you are rolling Cowork out across an organization, our enterprise deployment guide covers the related admin settings, and our security overview explains how Cowork handles files and connected apps.

How to get the most from Fable 5.1 in Cowork

1. Start on Opus 5.5. Run the task once on the cheaper model, and switch to Fable only if the result falls short.
2. Give it the whole job. Fable's advantage shows on long, multi-step work. A clear brief with the goal, sources, and the format of the finished file lets it run without stopping to ask.
3. Reuse context. Keep reference files in a project so repeated reads hit the cheap cache rate.
4. Watch usage in the first week. Check how much of your limit or credit balance a typical Fable task consumes before you schedule it to repeat.
5. Keep a human check. Even the strongest model makes mistakes in long outputs. Review figures and claims before results leave your team.

If a Fable session stalls or the model is missing from your picker, check your plan and credit balance first, then see our troubleshooting guide.

Frequently asked questions