Open Models Almanac
Methodsinference & servingcontext management

implementation detail · filed under inference & serving

Excluding prior thinking from multi-turn history

Omits prior assistant reasoning from subsequent conversation history, with tool-call turns retained where specified.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See context management for the methods filed beside it.

Also called No Thinking Content in History.

sources
6
models
3
labs adopt it
3
strongest
default

Documented in

Alibaba (Qwen)
Google DeepMind

Evidence

6 quotes from 6 sources, one card per model, each opening with its strongest.

How sources treat it
used 3default 3

Qwen3.5-397B-A17B

Alibaba (Qwen)adopted3 sources

In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content.

defaultinference servingin Qwen3.5-122B-A10BQwen Team
2 more from this model

In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content.

defaultinference servingin Qwen3.5-35B-A3BQwen Team

In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content.

usedsoftware implementationin Qwen3.5-397B-A17BQwen Team

Gemma 4 31B

Google DeepMindadopted2 sources

Thoughts from previous model turns must not be added before the next user turn begins, with the exception of tool call turns where thinking content should be preserved.

defaultinference servingin Gemma 4Google DeepMind
1 more from this model

In multi-turn conversations, the historical model output should only include the final response. Thoughts from previous model turns must not be added before the next user turn begins, with the exception of tool call turns where thinking content should be preserved.

usedinference servingin Gemma 4Google DeepMind

gpt-oss-120b

OpenAIadopted1 source

In general, you should drop any previous CoT content on subsequent sampling if the responses by the assistant ended in a message to the final channel.

usedinference servingin gpt-ossOpenAI
↑