Switching models mid-conversation: which reasoning survives?
Routing from Sonnet 5.5 to Haiku 5.5 and back, or falling back from GPT-6.2 to GPT-6.1? See which earlier reasoning the new model can read, which is dropped, and the caveats.
Runs in your browser. Nothing you paste is uploaded.
Full switch table (Claude API)
| Model you move from | To Claude Fable 5.2 / Mythos 5.2 | To Claude Fable 5.1 / Mythos 5.1 | To Claude Opus 5.5 | To Claude Sonnet 5.5 | To Claude Haiku 5.5 |
|---|---|---|---|---|---|
| Claude Haiku 5.5 | ✅ reads | — dropped | ✅ reads | ✅ reads | ✅ reads |
| Claude Sonnet 5.5 | ✅ reads | — dropped | ✅ reads | ✅ reads | — dropped |
| Claude Opus 5 | ✅ reads | ✅ reads | ✅ reads | — dropped | — dropped |
| Claude Opus 5.5 | ✅ reads | ✅ reads | ✅ reads | — dropped | — dropped |
| Claude Fable 5 / Mythos 5 | ✅ reads | ✅ reads | — dropped | — dropped | — dropped |
| Claude Fable 5.1 / Mythos 5.1 | ✅ reads | ✅ reads | — dropped | — dropped | — dropped |
| Claude Fable 5.2 / Mythos 5.2 | ✅ reads | — dropped | — dropped | — dropped | — dropped |
Every Claude model also reads blocks from Claude Opus 4.8, Claude Sonnet 5, Claude Haiku 4.5 and earlier Opus, Sonnet and Haiku models. Claude Opus 5 reads only itself and those; Claude Fable 5 and Mythos 5 read each other, Mythos Preview, Opus 5 and those.
Test-mode tip from Anthropic's docs: fail only on an input_transformations entry whose reason is prefix_binding_mismatch. Entries with model_binding_mismatch or organization_binding_mismatch are expected after a switch.
9 reason codes, 4 error strings and 22 fixes, each quoted from an official Anthropic or OpenAI page (last checked 2026-11-20). Reason codes · Error messages
How to use it
- Add the models the conversation used so far, oldest first. Or click Read models from my requests to take them from the Prefix check tab.
- Pick the model you’re switching to, and set the platform,
reasoning.contextand account options that apply. - Read one row per earlier model, then the caveats. Copy the share link: it holds only model IDs and options.
How the switch check works
For Claude, it reads Anthropic’s table of which models read which models’ thinking on the Claude API, plus the “earlier models” rules. A pair in neither table is shown as unknown. Off the Claude API, only what the docs confirm for that platform is used. Ticking the account box drops blocks from the account-bound models (Fable 5.2, Mythos 5.2, Sonnet 5.5 and Haiku 5.5).
For OpenAI, it uses the model family from the ID (gpt-6.2-astra → gpt-6.2): a different family drops the reasoning, the same family depends on a capability order OpenAI doesn’t publish. Across providers, neither provider’s docs describe reading the other’s reasoning.
How we check this
Every outcome, reason code, error string and fix is data in reasoning_rules.json with a short verbatim quote, its source and the date we checked it. The examples on this page are the same fixtures the tests run. Anthropic pages were last re-read on 2026-11-20; OpenAI pages on 2026-11-19.
- Change tools or instructions without failing the prefix check (Anthropic)
- Fix a thinking block bound to a different conversation (Anthropic)
- Mid-conversation system messages and tool changes (Anthropic)
- Prefix checks on thinking blocks (Anthropic)
- Preserved thinking (Anthropic)
- Claude Platform release notes (November 9, 2026) (Anthropic)
- Keep the prefix unchanged (OpenAI)
- Latest model (GPT-6.2) (OpenAI)
- Prompt caching (OpenAI)
- Responses API reference (OpenAI)
- Reasoning models (OpenAI)
- Structured outputs (OpenAI)
- Tool search (OpenAI)
Unconfirmed in the docs, so shown as "unconfirmed" or "unknown": whether a string and a text-block array count as the same content on Claude; how the 400 error's position maps when system or same-role messages merge; OpenAI's capability order within a family; reasoning.context defaults on GPT-6.1 Sol and GPT-6.2; whether tools, text.format and reasoning.effort changes remove every reasoning item on GPT-6.2; and switch rules off the Claude API.
FAQ
Does switching Claude models cause an error?
No. A thinking block that the new model or account can’t use never causes an error. The API drops it for that request and doesn’t bill it, and Claude answers without that reasoning.
Can any other model read Claude Fable 5.2's thinking?
No. Per Anthropic’s November 9, 2026 release notes, no other model reads thinking blocks from Claude Fable 5.2 or Claude Mythos 5.2. Fable 5.2 and Mythos 5.2 do read other models’ blocks on the Claude API and on Google-managed Google Cloud.
Should I keep a separate history per model?
No. Keep sending the whole conversation. Reasoning is lost permanently only if your code removes it; back on a model that can read the blocks, they’re read again.
What about OpenAI model switches?
Persisted reasoning can be reused only within the same model family, and moving to a less capable model in the family drops it too (model_mismatch). OpenAI doesn’t publish the capability order of astra, sol, terra and luna, so a same-family switch shows both outcomes.