Kimi-K2.5 (Azure OpenAI) returns responses with:
- "reasoning_content": "The..." — the actual model output goes here
- "content": null — empty
- "finish_reason": "length" — it uses all tokens on reasoning
CLASP translates content → Anthropic text block, but content is null, so the response is empty. The reasoning_content field is silently dropped.
Fix options (all in CLASP):
- CLASP should merge reasoning_content into content when translating OpenAI → Anthropic format
- Or CLASP should send a parameter to disable Kimi-K2.5's reasoning mode (Azure API may support this)
Kimi-K2.5 (Azure OpenAI) returns responses with:
CLASP translates content → Anthropic text block, but content is null, so the response is empty. The reasoning_content field is silently dropped.
Fix options (all in CLASP):