Claude Fable 5.1 and Claude Mythos 5.1

TL;DR Highlight

Fable 5.1 cuts cache read pricing, making typical workloads about 25% cheaper than Fable 5

Who Should Read

A backend developer running coding agents or long-running automation on the Claude API who is weighing cost against model choice

Core Mechanics

Fable 5.1 and Mythos 5.1 are the same model
- They differ only in the level of safeguards.
- Fable 5.1 is generally available, while Mythos 5.1 is limited to trusted access programs.
Cheaper cache reads cut costs
- Billed by token, typical workloads are estimated to cost about 25% less than Fable 5.
- Highly agentic work can save up to about 45%.
Low and Medium effort already match Fable 5
- At Low or Medium effort, Fable 5.1 reaches results similar to or better than Fable 5 at much lower cost.
- Default effort is High in Claude Code and Medium in Claude Cowork and on Claude.ai.
Benchmarks beat Fable 5 and Opus 5
- Terminal-Bench-Science 0.1 is 52.6%, versus Fable 5 at 24.7%, Opus 5 at 29.0%, and GPT-5.6 Sol at 22.4%.
- CursorBench 3.2.0 is 73.4%, versus Fable 5 at 70.5%, Opus 5 at 70.0%, and GPT-5.6 Sol at 67.2%.
- AutomationBench is 31.4%, versus Fable 5 at 17.1%, Opus 5 at 26.9%, and GPT-5.6 Sol at 19.6%.

Community Reactions

How to Apply

Run some Opus 5 traffic on Fable 5.1 Medium today
- Pick one task you kept on Opus, such as code review, and run both models on the same inputs.
- Record tokens, cost per task, and output quality side by side.
Plot your own accuracy-versus-cost curve by effort
- Run your eval set at low, medium, and high and measure mean cost per task.
- Set the lowest effort with good enough quality as your default.
Recalculate agent costs by cache read share
- Check the share of cache read tokens in your billing data.
- Move the long-running agent jobs with the highest share to Fable 5.1 first and measure the savings.
Check eligibility for EFS or zero data retention if you have retention requirements
- Ask your account contact whether your organization qualifies for zero data retention.
- Plan adoption around the EFS availability timeline.

Try It

copy & run

# Paste into Claude Code or Claude: model switch cost comparison plan (a simplified version based on the article)

Our service currently uses {{current model}} for {{task type}}. Monthly token usage is {{input tokens}} input, {{output tokens}} output, and {{cache read tokens}} cache reads.

We are deciding whether to move to Fable 5.1. Draft an experiment plan with these steps.
1. Criteria for picking 20 representative tasks
2. Metrics to measure at low, medium, and high effort (accuracy, cost per task, duration)
3. A table calculating how the cache read price cut affects our costs
4. A decision rule for switching (for example, if quality is equal, choose the cheaper option)

Next step: after running the 20 tasks, paste the results table and ask which effort level fits best.

Terminology

Cache readThe model rereads input that was already processed and stored. It is like resuming from a bookmark instead of reading the book from the start.

EffortA setting for how long and how hard the model works on a task. Lower is cheaper and faster, higher is pricier and more thorough.

False positiveA harmless request that gets wrongly blocked as risky. It is like a spam filter putting a legitimate email in the spam folder.

Zero data retentionA policy where the provider does not store the inputs and outputs you send through the API.

Enterprise Frontier Safeguards (EFS)A new approach that stores data in cloud infrastructure controlled by the customer rather than Anthropic while keeping safeguards against adversarial use.