Claude Fable 5.1 is here: pricing, upgrades, and the Mythos difference

Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. Fable 5.1 is generally available, while Mythos 5.1 is limited to trusted access programs. Anthropic says they share the same underlying model and differ in the scope of their safeguards.[2]

The short version: Fable 5.1 is designed for coding, research, and multi-step agent jobs that run for hours rather than one-off prompts. The most useful changes are not limited to benchmark gains. Cheaper cache reads, fewer requests for supervision during long tasks, clearer reports when the model gets stuck, and more natural writing can all affect day-to-day use.[1][3]

This article is based on Anthropic's launch material and product documentation. I have not yet run the model through an independent test suite, so the performance figures below should be read as vendor-reported results.

Quick summary

• Fable 5.1 is live in Claude Code and the Claude Platform.[1]
• Pro, Max, Team, and Enterprise users can access it. The API model ID is claude-fable-5-1.[3]
• API pricing remains $10 per million input tokens and $50 per million output tokens, the same as Fable 5.[3]
• Cache reads now cost $0.25 per million tokens, which Anthropic describes as a 75% reduction.[3]
• Anthropic estimates about 25% lower cost for typical workloads and up to roughly 45% for highly agentic work.[2][3]
• Mythos 5.1 uses the same underlying model with broader access for cybersecurity and life-science work, but it is not generally available.[2]

The list price is unchanged, but long tasks can cost less

Fable 5.1 keeps Fable 5's standard API pricing: $10 per million input tokens and $50 per million output tokens. The meaningful cut is in caching. Reading previously processed context now costs $0.25 per million tokens, 75% less than before.[1][3]

That matters more for agents that repeatedly load large repositories, instructions, and documents than for a short chat. Systems that reuse the same context throughout a long job benefit most from cache pricing. Anthropic's estimates of 25% savings for typical workloads and up to 45% for highly agentic work depend on that usage pattern.[2]

Fable is still not a cheap default for every request. A smaller model may remain the better choice for simple summaries, classification, and short code edits. Fable 5.1's cost case is strongest when a task needs Fable-level capability and can reuse substantial context.

What changed in long-running work?

The ClaudeDevs announcement says Fable 5.1 can get further into a long task before asking for input, does a better job of saying when it is stuck, and writes more naturally.[1] Anthropic's product page positions it for work that lasts hours and crosses several applications: processing a backlog, operating a browser, running as a managed agent, or making a codebase-wide change.[3]

The useful measure is not simply how many minutes an agent runs unattended. Does it keep reliable notes, recognize a failed step, recover sensibly, and stop at the right boundary? Anthropic says Fable 5.1 is better at finding root causes, writing its own tests, and using vision to compare an output with the original goal.[3] Those are strong claims, and real projects still need acceptance tests and human approval.

Claude Code uses High effort by default for Fable 5.1, while Claude.ai and Cowork default to Medium.[2] Teams watching cost should pin the effort level and compare models on the same tasks rather than looking at the model name alone.

What do the benchmarks say?

In Anthropic's results, Fable 5.1 reaches 52.6% on Terminal-Bench-Science 0.1, compared with 24.7% for Fable 5. AutomationBench rises from 17.1% to 31.4%, while CursorBench 3.2.0 moves from 70.5% to 73.4%.[2]

There are two reasons to be cautious. The model vendor published the tests, so teams should validate the results independently on their own task sets. Anthropic also says production safeguards reduced scores on some tasks and reports a standard error of 3.5 to 4.5 points per model for Terminal-Bench-Science.[2]

A small internal suite is more useful than a single leaderboard number. Five real coding tasks, two long research jobs, and one failure scenario will tell most teams more about a migration than the launch table will.

Fable 5.1 versus Mythos 5.1

Anthropic describes Fable 5.1 and Mythos 5.1 as the same underlying model. Fable is generally available with stricter controls in cybersecurity, biology, and chemistry. Mythos 5.1 is limited to vetted organizations through trusted access programs.[2]

Some cybersecurity and biology requests sent to Fable are routed to less capable Opus models. Anthropic says Fable 5.1 can now be used more broadly for defensive vulnerability discovery, while exploit generation and several dual-use tasks remain restricted or rerouted.[2][3]

API teams need to account for this behavior. The product page says fallback must be configured for API use.[3] Selecting one model ID does not necessarily mean every request will be answered by that model.

Do not skip the data-retention policy

Fable uses 30-day data retention for safety monitoring by default.[3] Anthropic says its new Enterprise Frontier Safeguards system will let eligible enterprise customers keep data in customer-controlled cloud infrastructure. The rollout is planned in phases beginning in fall 2026, with zero data retention available to eligible customers in the meantime.[2][3]

For teams handling proprietary code or company data, this may matter more than the benchmark table. Confirm the product surface, contract, region, and fallback data path in writing before migrating.

If you are considering access through AWS, the Claude Platform on AWS versus Amazon Bedrock guide covers the data-boundary and compliance differences. The local versus cloud coding-agent guide handles the separate question of where the agent should run.

Who should switch, and who should wait?

If you already use Fable 5 for long coding or research jobs, testing Fable 5.1 on the same task set makes sense. Standard token pricing is unchanged, cache reads are cheaper, and low or medium effort may offer a better cost-performance balance.[2][3]

If your workload is mostly short questions, simple content generation, or small code edits, making Fable 5.1 the default for everything may be wasteful. Keep a smaller model for routine work and call Fable for long, high-value tasks.

Enterprise teams should start with retention and safeguard routing. A more capable model is not automatically an acceptable destination for company data.