Reasoning removal reason codes
Every reason Claude and OpenAI give in input_transformations for dropping (or letting through) earlier reasoning. Paste a response into the decoder to read yours.
Anthropic (Claude)
| Entry | Meaning | Yours to fix? |
|---|---|---|
thinking_dropped prefix_binding_mismatch | Something before this thinking block (system, tools or an earlier message) changed after Claude produced it, so the API dropped the block. That happens with prefix_mismatch_behavior: "drop_block". | Yes |
thinking_mismatch_allowed prefix_binding_mismatch | The thinking block failed the prefix check, but with the setting unset this model let it through: Claude read it unchanged and you were billed for it. It’s reasoning about content that has since changed. | Yes |
thinking_dropped model_binding_mismatch | You switched models mid-conversation and the new model can’t read the old model’s thinking. The API drops it for this request only, with no error and no charge. | No |
thinking_dropped organization_binding_mismatch | The thinking block came from a different account (organization, AWS account or Google Cloud project), so the API dropped it. | No |
OpenAI
| Entry | Meaning | Yours to fix? |
|---|---|---|
prefix_mismatch | Content before this reasoning item was inserted or edited, so GPT-6.2 or later removed the item. The request still succeeded. | Yes |
end_user_mismatch | The request that reused this reasoning item had a different safety_identifier from the one that created it (or only one of them had one). | Yes |
model_mismatch | You switched to a different model family, or to a less capable model in the same family, so the earlier reasoning items were removed. | Usually no |
organization_mismatch | The reasoning item was created in another organization (or another workspace under a different user), so it was removed. | No |
turn_boundary | The model used only current-turn reasoning, so a reasoning item from an earlier turn wasn’t used. | Maybe |
Anthropic's docs: "Ignore an entry whose type or reason you don't recognize." The research behind this tool mentioned an Anthropic end_user_binding_mismatch; it isn't in Anthropic's docs, so it isn't listed.
Last checked 2026-11-20 against the official docs.