Unverified60% confidenceFactExact time
On DeepSeek's own API, tool calls in thinking mode must send back the reasoning_content of the historical assistant message in every turn, otherwise a 400 is returned.
2
Sources
60%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDeepSeek Harness adopts an "append-only, no tampering" session log mechanism, achieving 100% traceability for every run.71% similarVerifiedDeepSeek-Harness writes all content in a conversation—including system prompts, chain-of-thought, tool calls, and everything else—into an append-only session log.70% similarUnverifiedDeepSeek introduced a 'controllable reasoning intensity' integer parameter, allowing users to set a value from 1 to 100 in API requests to trade off between reasoning cost and accuracy.69% similarUnverifiedThe DeepSeek API plays two roles in this project: a dialogue generation engine and a memory value evaluator.68% similarUnverifiedDeepSeek Harness supports exporting complete conversations in JSON format and viewing metrics such as the number of context injections, reasoning processes, and token consumption.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067652API
curl https://kongchang.com/api/v1/knowledge/claims/1067652MCP
get_claim(id=1067652)