Claude Fable 5.2 Shipped Nov 9. If Your Bot Forces Tool Choice, Check This First
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, with lower prices and a faster output mode. Two changes matter more than the headline numbers for a bot that calls tools: forcing a specific tool now returns a 400 error instead of running, and the default reasoning effort jumped a level higher than Opus 5.5's default. Test both before you point production traffic at the new model ID.
AI-assisted editorial: researched and drafted with AI using the primary sources linked below. Check the cited evidence and any stated test scope. Illustrative examples are not measured product results. Editorial policy.
Key takeaways
- On claude-fable-5-2, tool_choice values of "any" or a named "tool" return a 400 invalid_request_error — bots that force a specific tool call will break, not just behave differently.
- Fable 5.2's default effort is "high," one level above Opus 5.5's "medium" default, so a request that omits effort runs more tokens and costs more than before.
- Pricing dropped 20% to $8/$40 per million tokens, but a forced-tool-choice bot pays more in broken calls than it saves in token price if it migrates without this check.
What Anthropic shipped on November 9
Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, pitching both as upgrades aimed at the hardest, most open-ended technical work rather than everyday coding or chat tasks. Fable 5.2 is available broadly across Anthropic's platforms, while Mythos 5.2 stays restricted to organizations approved through Anthropic's verification programs for cybersecurity and life-sciences work.
The models share one underlying checkpoint and the same specs: a 1M token context window by default and up to 128k output tokens per request. Pricing fell to $8 per million input tokens and $40 per million output tokens, down from $10 and $50 on the 5.1 generation, with an optional fast mode that speeds generation by up to 2.5x at double the standard token price.
Sources: Introducing Claude Fable 5.2 and Claude Mythos 5.2 \ Anthropic, What's new in Claude Fable 5.2.
The change that breaks a forced tool call
If your bot pipeline sets tool_choice to force a specific tool, this is the line to check before migrating. Anthropic's migration guide documents a breaking change already present since the Fable 5 generation and still true on Fable 5.2: Claude Fable 5 accepts tool_choice values of auto, none, any, and tool, but on claude-fable-5-2, the any and named-tool forms return a 400 invalid_request_error, and the token-counting endpoint rejects them the same way.
This is not a behavior change where the model ignores the forced tool and picks its own path — the request fails outright. Any bot that relies on forced tool_choice to guarantee a specific function gets called (a common pattern for structured extraction or a mandatory lookup step) needs a code path change, not just a prompt tweak, before it can run on Fable 5.2 or Mythos 5.2.
A related failure shows up before a single tool call even runs: bots that pre-check token usage against a budget by calling Anthropic's token-counting endpoint will see that endpoint reject the same tool_choice values of any or a named tool with the identical 400 invalid_request_error, per the migration guide. An operator who only tested the live messages endpoint may miss this and get surprised when a pre-flight budgeting step starts failing on Fable 5.2 even though the main call path was already fixed. The detection signal is the same invalid_request_error body pointing at tool_choice, just surfacing on a different endpoint. Recovery is identical to the main fix: switch any forced any or named-tool value to auto or none wherever token counting is called, not just where the tool call itself executes.
Sources: Claude Fable 5.2 and Claude Mythos 5.2 migration guide.
The default effort jump raises cost and latency quietly
A second migration note matters even if your bot never forces a tool. Anthropic's guide states that Claude Fable 5.2's default is high, where Claude Opus 5.5's default is medium, so a request that omits the effort parameter runs one level higher than before. The guide's own advice is to set effort explicitly rather than inherit whatever the new default happens to be, then re-run your effort sweep on your own evals and re-baseline cost and latency.
That matters because a team that copies an Opus 5.5 call pattern onto Fable 5.2 without setting effort will see both higher token spend and longer wall-clock time per call, and it will look like the new model is simply slower or pricier rather than a parameter default that moved underneath them.
The practical symptom of the effort default change is a billing or latency anomaly rather than an outright error, which makes it slower to notice. An operator watching per-request token counts or a cost dashboard will see average output tokens and wall-clock time per call rise after a model-string swap to Fable 5.2, with no code change to explain it, because omitted effort now runs at high instead of the medium default carried over from Opus 5.5, as the migration guide states. The fix the guide recommends is to stop relying on the default entirely: set effort explicitly on every call, then re-run an effort sweep against existing evals to find the lowest effort level that still meets quality bar, and re-baseline cost and latency against that explicit setting rather than against pre-migration numbers.
Sources: Claude Fable 5.2 and Claude Mythos 5.2 migration guide.
Illustrative example: a bot that forces a validation lookup
Say a small-team support bot runs on Fable 5.1 today and forces a single tool, verify_account, every time a user message contains an account number, using tool_choice: {type: "tool", name: "verify_account"}. This is an illustrative pattern, not a measured BotBento test. Under the current Fable 5.1 model this call succeeds and the model has no choice but to invoke that tool.
Swap the model string to claude-fable-5-2 without touching anything else, and that same request now returns a 400 invalid_request_error instead of a tool call. If the bot's error handling treats any 400 as a transient failure and retries with backoff, the account-verification step silently fails every time until someone reads the error body and finds invalid_request_error pointing at tool_choice. The fix is to switch the forcing logic to a system-prompt instruction or an auto tool_choice with a tightly scoped tool list, then validate with the model's own evals before trusting it in production — exactly the kind of deeper precedence check this site's forced-tool-choice guide covers for tool_choice generally.
Sources: Claude Fable 5.2 and Claude Mythos 5.2 migration guide.
Decision: test this week, migrate only after two checks pass
Test Fable 5.2 this week if your bot does long-horizon agentic work, deep multistep research, or document-scale analysis where Anthropic positions the model, and if your current Opus 5.5 evals are falling short on complexity. Don't flip production traffic until you've run these two checks.
First, grep your codebase for tool_choice settings of any or a named tool pointed at a Fable or Mythos model ID, and change any you find before testing — they will fail outright, not degrade. Second, set the effort parameter explicitly on every Fable 5.2 call instead of relying on the default, then re-run your cost and latency baseline, because the default moved one level above what Opus 5.5 users are used to. Both checks come straight from Anthropic's own migration guide, not from inference about how the model behaves.
- Check 1: search for forced tool_choice (any / named tool) on any Fable or Mythos 5.x model string — it now returns a 400, not a different tool pick
- Check 2: set effort explicitly on Fable 5.2 calls; the default is high, one level above Opus 5.5's medium default
- If neither check applies to your bot, the 20% price drop and 1M context window make this a low-risk model to test this week
Sources: Claude Fable 5.2 and Claude Mythos 5.2 migration guide, Pricing - Claude Platform Docs.
Primary sources
Sources checked 2026-11-10. Standards and product documentation can change; follow the linked version when implementing.
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 \ Anthropic — Anthropic
- Claude Fable 5.2 and Claude Mythos 5.2 migration guide — Anthropic
- What's new in Claude Fable 5.2 — Anthropic
- Pricing - Claude Platform Docs — Anthropic
BotBento is in development. Suggest a correction.