Updated November 10, 2026
What is Claude Fable 5.2?
Claude Fable 5.2 is Anthropic's premium model for demanding reasoning and long-horizon agentic work. Released on November 9, 2026, it is designed for tasks that need sustained investigation, tool use and a finished technical deliverable rather than a quick conversational answer. Anthropic lists a one-million-token context window, a 128,000-token maximum output and text-and-image input with text output. The API model ID is claude-fable-5-2. Its place in the lineup is important: Anthropic still recommends starting with Opus 5.5 for most coding and knowledge tasks, then moving up when the task or evaluation calls for it.
Choose Fable 5.2 for difficult engineering, research and multi-step work when its extra capability justifies premium token costs. For most everyday coding and knowledge tasks, Anthropic recommends starting with Opus 5.5 and moving up when the task demands it.
Official model and release documentation
Specification | Verified detail |
|---|---|
Provider | Anthropic |
Model ID | claude-fable-5-2 |
Access | Active public model; platform and account availability apply |
License | Proprietary hosted model |
Context window | 1M tokens |
Maximum output | 128K tokens |
Modalities | Text and images in; text out |
Thinking | Adaptive, always on |
Default effort | High |
Knowledge cutoff | June 2026 |
Standard input/output | $8 / $40 per million tokens |
What changed and how to approach migration
Fable 5.2 follows Fable 5.1 with improvements that Anthropic reports in coding, reasoning, vision and communication. The release announcement says token generation can be up to 30% faster than the previous Fable and Mythos generation, and adds an optional paid fast mode. These are provider claims under the provider's conditions, not our own latency measurements. The practical migration question is whether the new model completes a hard task with fewer retries, less steering or a better final artifact. A faster token stream alone does not establish a faster end-to-end workflow.
Area | What changed |
|---|---|
Predecessor | Claude Fable 5.1 |
Positioning | More demanding technical and long-running agent tasks |
Generation speed | Up to 30% faster, according to Anthropic |
Optional fast mode | Up to 2.5 times token generation speed; separate price |
Safeguards | Cyber and biology safeguards remain in the public Fable model |
Capabilities and practical use cases
A useful Fable pilot should resemble a real engineering assignment: inspect an unfamiliar repository, explain the architecture, propose a bounded change, implement it and deliver a reviewable result. For research, ask for a sourced synthesis with explicit unresolved questions rather than an unsupported final answer. Long context can keep more source material available, but the model still needs a clear definition of success. Anthropic's examples include software migrations, analysis and scientific work. Those examples support considering the model for ambitious tasks; they do not establish that every team can reproduce the same outcomes with an ordinary prompt.
Workflow | Practical fit |
|---|---|
Repository migrations | Maintain invariants while changing multiple components |
Difficult debugging | Investigate competing explanations and test a repair |
Deep research | Carry sources and evidence through to a final deliverable |
Visual analysis | Interpret screenshots and diagrams alongside instructions |
Agent orchestration | Coordinate smaller-model work with explicit review points |
Document production | Finish a technical document, spreadsheet or presentation workflow |
Benchmarks and evidence
Anthropic reports 49.1% on FrontierCode 2.0 at xhigh effort and 69.9% on FrontierSWE v2. These evaluations support considering Fable 5.2 for difficult engineering, but they do not measure the same task or establish a universal success rate. The release also distinguishes Fable results with safeguards from some specialist Mythos results. A shared underlying model does not make those test configurations interchangeable. For a practical comparison, measure whether the model produces a correct, maintainable result on the repositories and research tasks your team actually handles.
Evidence | Reported result | How to interpret it |
|---|---|---|
FrontierCode 2.0 | 49.1% | Anthropic-reported; xhigh effort |
FrontierSWE v2 | 69.9% | Anthropic-reported evaluation |
OSWorld 2.1 | 89.3% | Offline partial-credit set in release table |
Public safeguards | Enabled unless otherwise noted | Read the system card and benchmark footnotes |
Pricing and total cost
Standard Claude API pricing is $8 per million input tokens and $40 per million output tokens. Anthropic lists $0.20 per million cache-read tokens, $10 for a five-minute cache write and $16 for a one-hour cache write. The announcement prices optional fast mode at $16 input and $80 output. As a simple arithmetic illustration, one million uncached input tokens plus 100,000 output tokens costs $12 at the standard base rates. That illustration excludes additional tool or platform charges and does not predict the amount of reasoning or output a task will consume.
Cost component | Published term or planning boundary |
|---|---|
Standard input | $8 / million tokens |
Standard output | $40 / million tokens |
Cache read | $0.20 / million tokens |
5-minute cache write | $10 / million tokens |
1-hour cache write | $16 / million tokens |
Optional fast mode input/output | $16 / $80 per million tokens |
Illustrative 1M input + 100K output | $12 base token cost, standard mode |
Limitations and deployment tradeoffs
The main tradeoff is paying for a model intended for unusually hard work when a cheaper model would already meet the acceptance threshold. Fable's cyber and biology safeguards also make it a different operational choice from verified-access Mythos. A one-million-token context window does not guarantee perfect recall, and a 128K output ceiling does not make a very long answer desirable. Test whether the model finds the relevant evidence and produces a concise, maintainable result. Public availability should not be confused with identical access, quotas or feature support across every hosting platform.
Consideration | What to check |
|---|---|
Premium cost | Route routine extraction and classification elsewhere |
Long-running sessions | Limit total tokens, steps and elapsed time |
Large context | Test instruction retention and evidence retrieval |
Specialist safeguards | Use the documented access path for legitimate specialist work |
Provider benchmark gaps | Repeat representative tasks in your own environment |
How to evaluate Claude Fable 5.2 for your workload
Start with a small, representative set of multi-step engineering and research tasks, including ordinary cases, difficult edge cases and tasks with an objectively checkable answer. For this model, a useful pilot includes a repository migration with regression checks and a sourced research deliverable. Keep the prompts, available tools, source material and acceptance criteria consistent across candidates. Record the exact model ID, reasoning configuration, date and any provider-specific access program so the comparison can be repeated later.
Measure completed-task quality rather than the fluency of the response. Ask a reviewer to identify unsupported claims, incorrect changes, missing evidence and avoidable manual repair. An agent that stops honestly when evidence is insufficient can be more useful than one that produces an impressive-looking but invalid artifact. Where tools are involved, track invalid arguments, failed calls and recovery behavior separately from the final answer. A single successful demonstration is a reason to continue testing, not proof of a dependable service level.
Pilot measure | What to record |
|---|---|
Task completion | Was the requested artifact actually completed and checked? |
Evidence quality | Can a reviewer trace decisive claims to sources or tests? |
Correction burden | How much expert time was needed after the model finished? |
Latency | Measure end-to-end time, not just first-token speed |
Total cost | Include tokens, tools, retries and human review |
Failure handling | Did it report uncertainty and recover without inventing success? |
Use a staged rollout only after the pilot meets the target you set in advance. Preserve a known baseline and a way to compare new requests with the previous configuration. If the model, tool interface, access policy or price changes, rerun the affected cases instead of assuming an old result still applies. The purpose of evaluation is a defensible choice for your actual work, not a claim that one model is universally best.
Who should use it, and what should you compare?
Fable is a sensible shortlist candidate for teams whose difficult tasks fail on their normal daily-driver model. It is less compelling as the default for every support reply, classification step or small edit. Treat it as a model to escalate to after identifying the failure modes that matter. Compare it with Opus 5.5 at the effort settings you would really use, not only with an artificially weak baseline. For tasks with clear outputs and inexpensive automated checks, a stronger model can be valuable even when its per-token rate is higher.
Alternative | Why compare it | Decision boundary |
|---|---|---|
Claude Opus 5.5 | Daily-driver coding and knowledge work | Start here unless the harder task needs Fable |
Claude Sonnet 5.5 | Speed and cost balance | Useful for routine parts of an agent workflow |
GPT-6 Astra | Cross-provider frontier comparison | Evaluate with the same tools and success criteria |
Claude Mythos 5.2 | Verified specialist use cases | Access and safeguards differ; not a public bypass |
Frequently asked questions
Is Fable 5.2 the best model for every task?
No model is best for every task. Anthropic recommends Opus 5.5 as a starting point for most coding and knowledge work. The better purchase depends on task difficulty, cost, latency and the amount of human correction required.
How is it different from Mythos 5.2?
Anthropic describes the same underlying model, with public Fable safeguards and specialist Mythos access through verification programs. Do not assume a Fable benchmark or access entitlement automatically transfers to Mythos.
Does the million-token window guarantee perfect recall?
No. It is a capacity limit. Test whether the model can find the decisive detail, distinguish relevant evidence from distracting material and retain your instructions at the lengths your application actually uses.
Should an existing Fable 5.1 workflow switch immediately?
Run a controlled comparison first. Preserve tools, task inputs and acceptance criteria, then compare completed-task quality, cost and wall-clock time. Review the provider's migration guidance before changing the production model ID.
Related model guides
Compare the full AI model directory
Sources and editorial method
Fable 5.2 specifications and pricing
Specifications, access conditions and prices were checked against the linked provider sources on November 10, 2026. Provider-reported benchmarks are attributed and are not presented as independently reproduced tests. Missing specifications remain unverified rather than being inferred from a related model. Practical workflow advice is an evaluation recommendation, not a claim that we ran new hands-on trials. Recheck official terms before purchase or deployment.