Key takeaways
- Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. The generally available Claude Platform ID is claude-fable-5-2; Mythos access requires verification.
- Standard pricing is $8 per million input tokens and $40 per million output tokens. Optional fast mode costs $16 input and $80 output; Anthropic advertises up to 2.5 times faster token generation in that mode.
- Anthropic says ordinary generation is up to 30% faster than Fable 5.1 and reports stronger coding, reasoning and vision results. These are vendor claims, not AccessAllGPT measurements.
- AWS confirms Amazon Bedrock and Claude Platform on AWS access. Its example uses global.anthropic.claude-fable-5-2; Bedrock users must review the aws_review retention setting before invocation.
- OpenRouter lists a 1,000,000-token context and 128,000-token maximum completion for its Fable 5.2 routes. Those are registry limits, not a guarantee of long-context accuracy or identical limits on every cloud.
- Start with a bounded comparison against your existing model. Anthropic still recommends Opus 5.5 for most everyday coding and knowledge work, reserving Fable for harder tasks.
What launched on November 9
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, with improvements it describes in coding, reasoning and vision. Claude Fable 5.2 is generally available, including through AWS, Google Cloud and Microsoft Azure. Developers using the Claude Platform can request claude-fable-5-2. Standard pricing is $8 per million input tokens and $40 per million output tokens; an optional fast mode costs $16 and $80 respectively.
The two names do not mean two unrelated base models. Anthropic says Fable and Mythos share the same underlying model, with different safeguards and access conditions. Mythos 5.2 is restricted to approved organizations through verification programs. Fable is the relevant starting point for general-purpose builders; this briefing does not cover biological applications or recommend circumventing safeguards.
The practical news is a less expensive successor to Fable 5.1 with a new speed option, not a reason to replace every smaller model. Anthropic itself recommends starting most coding and knowledge tasks with Opus 5.5 and moving upward to Fable when the task warrants it.
Standard mode gets cheaper; fast mode is a separate purchase
The new standard rates are below Fable 5.1’s $10 input and $50 output per million tokens, the predecessor rates tracked in AccessAllGPT’s price directory. Anthropic also says Fable 5.2 generates outputs up to 30% faster than Fable 5.1. That is a claim about token generation, not a measured reduction in end-to-end application latency.
Fast mode doubles the new standard token rates to $16 input and $80 output per million tokens. Anthropic advertises up to 2.5 times faster token generation. Its launch post lists fast-mode availability in Claude Code, the Claude Platform, Claude Models on Google Cloud under Anthropic-managed and Google-managed options, and Microsoft Foundry under Anthropic-hosted access. Do not assume the standard AWS launch establishes fast-mode support on Bedrock.
For a long-running agent, compare cost per accepted result, not just tokens per second. Reasoning effort, retries, tool latency, review time and the number of subagents can outweigh a lower token price. Fast mode is worth evaluating where waiting is expensive; it is not automatically the economical default.
Coding gains are promising, but the published scores are vendor evaluations
Anthropic reports 49.1% on FrontierCode 2.0 at Xhigh effort for Fable 5.2, compared with 32.4% for Fable 5.1 at Xhigh. It reports 69.9% versus 56.3% on FrontierSWE v2. These are vendor-reported scores from the launch table, not independent AccessAllGPT tests or forecasts for a reader’s repository.
Effort changes the comparison. Anthropic says Fable 5.2 at High effort scored 45.3% on FrontierCode, close to Opus 5.5’s 45.5% best result, at roughly half the cost per task in its evaluation. That is a useful hypothesis to test with a fixed harness, rather than evidence that every Fable request is cheaper than an Opus request.
The announcement also describes large software migrations and an operating-system prototype coordinated across many agents. Those examples illustrate what Anthropic and its early testers attempted; they do not establish production reliability, secure defaults or acceptable review overhead for other teams. Keep permissions narrow and require real test evidence before accepting generated changes.
The route determines the API contract and retention obligations
AWS independently confirms availability on Amazon Bedrock and Claude Platform on AWS. Its Bedrock example invokes global.anthropic.claude-fable-5-2, and the post lists US Geo CRIS and Global CRIS inference profiles. AWS describes Messages API access through the Anthropic SDK as well as Invoke and Converse APIs. Confirm regional support and account permissions before treating a model-name substitution as a completed migration.
AWS calls Fable 5.2 a Covered Model: default use requires up to 30 days of retention and human safety review by Amazon personnel within the AWS boundary. Its instructions say to set the aws_review data-retention mode on the account or project before invocation. Eligible zero-data-retention arrangements have separate conditions; AWS describes internal-use eligibility through at least June 30, 2027 while Enterprise Frontier Safeguards is being introduced.
Anthropic describes Enterprise Frontier Safeguards as a private beta that runs automated monitoring on prompts and responses stored in customer-controlled infrastructure. Its own announcement describes 30-day safety-monitoring retention for other customers. Do not transfer an eligibility statement or privacy assumption from one route to another without checking the applicable agreement.
Anthropic also says preserved thinking now applies to all accounts. Review the linked integration documentation before changing how prior reasoning context is carried between requests. AWS’s migration guidance specifically calls out model IDs, adaptive thinking, deprecated settings and output_config.effort as things to check.
OpenRouter exposes standard and batch IDs with a million-token context
The public OpenRouter registry retrieved on November 10 lists anthropic/claude-fable-5.2 and anthropic/claude-fable-5.2:batch. Both expose a 1,000,000-token context, a 128,000-token maximum completion and text, image and file inputs with text output. These are OpenRouter registry specifications; this briefing does not assert that every Anthropic or cloud endpoint has the same limits.
The standard record lists $8 input and $40 output per million tokens. The batch record lists $4 input and $20 output and exposes /api/v1/batches. Both mark reasoning as mandatory and list low, medium, high, xhigh and max effort, with high as the registry default. Anthropic separately describes Medium as the default in Claude apps and Claude Code, so defaults should not be assumed to match across surfaces.
A registry entry is evidence of a listed route, not proof that our account can call it or that its advertised context performs well on every task. AccessAllGPT did not call either route, submit a batch, test tools or measure latency. No hands-on result is implied.
What builders should do next
For difficult coding work, run a small, fixed comparison against your existing model at explicit effort levels. Use representative issues with executable acceptance tests, retain raw outputs and tool traces, and count human corrections. Compare standard and fast mode only on a route that explicitly supports them; compare batch separately for work that can wait.
Before sending private code or documents, confirm retention, review access, data location and zero-retention eligibility for the exact service. Stop when a request is refused; do not change models or rephrase restricted material to bypass that decision. Treat a higher benchmark score as a reason to evaluate, not permission to expand an agent’s authority.
A short Fable 5.2 rollout record
Keep this alongside a bounded trial rather than making the launch itself a production approval.
Entries stay in this browser tab and are not submitted to AccessAllGPT. Blank responses are copied as [Unresolved].
Record the exact model ID, cloud, region, effort, token limits and standard, fast or batch mode.
Confirm retention, human review, customer-controlled monitoring and any contractual zero-retention eligibility.
Use fixed tasks and tests; record accepted results, regressions, corrections, end-to-end time and total billed usage.
Keep the current model as a baseline; approve only the workloads where quality, cost and governance improve, with a rollback owner.
Primary sources
Browse the publication-wide evidence index →
- Introducing Claude Fable 5.2 and Claude Mythos 5.2Anthropic · Reviewed: November 9 date; introduction; performance and efficiency; coding; safeguards; data privacy; availability · Retrieved · Supports: First-party announcement of the shared underlying model, general versus verified access, standard and fast pricing, vendor-reported coding scores, speed claims, cloud availability, preservation of thinking and retention arrangements.
- Introducing Claude Fable 5.2 on AWSAWS · Reviewed: November 9 date; data retention and Enterprise Frontier Safeguards; getting started; migration; availability · Retrieved · Supports: Confirms Bedrock and Claude Platform on AWS availability, global.anthropic.claude-fable-5-2 invocation, US Geo and Global CRIS, aws_review retention configuration and conditional zero-retention arrangements.
- OpenRouter Models API: Claude Fable 5.2 recordsOpenRouter · Reviewed: Standard and batch model IDs; context_length; top_provider; architecture; pricing; endpoints; reasoning defaults · Retrieved · Supports: Registry lists standard and batch Fable 5.2 routes, million-token context, 128,000-token maximum completion, modalities, standard and batch token rates, mandatory reasoning and high default effort. This is registry evidence, not an inference test.
Limitations
This is a source-backed launch briefing, not a benchmark, hands-on review or production certification. We did not call Fable or Mythos, test coding quality, fast mode, batch processing, context accuracy, SDK compatibility, preserved thinking or retention controls, inspect the full system card, verify every cloud’s region list or reproduce vendor speed and evaluation results. Public availability does not establish account eligibility, quotas or service-level guarantees. Registry prices, limits and defaults may change. Biological applications are outside this unattended article’s scope.
Disclosures
AccessAllGPT did not receive model access, credits, payment or a vendor briefing for this article. Anthropic, AWS and OpenRouter did not review or endorse it. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with these vendors. Publication-wide relationships are listed on the disclosures page.
