-
fix: read cache_write_tokens from OpenRouter API instead of hardcoding 0 (#9871)
frostbyte_neo 发布于
2026-03-20 22:46:29 +00:00 - fix: read cache_write_tokens from OpenRouter API instead of hardcoding 0
- Read prompt_tokens_details.cache_write_tokens from OpenRouter stream usage
chunks instead of hardcoding cacheWriteTokens to 0 - Read native_tokens_cache_write from generation endpoint fallback
- Replace fragile hardcoded model ID switch statement for cache_control blocks
with prefix-based matching (anthropic/, minimax/) so new models automatically
get prompt caching enabled - Add unit test verifying cache_write_tokens are correctly extracted
Co-authored-by: Ara arafat.da.khan@gmail.com
- fix: read cache_write_tokens from OpenRouter API instead of hardcoding 0
- Read prompt_tokens_details.cache_write_tokens from OpenRouter stream usage
chunks instead of hardcoding cacheWriteTokens to 0 - Read native_tokens_cache_write from generation endpoint fallback
- Replace fragile hardcoded model ID switch statement for cache_control blocks
with prefix-based matching (anthropic/, minimax/) so new models automatically
get prompt caching enabled - Add unit test verifying cache_write_tokens are correctly extracted
Co-authored-by: Ara arafat.da.khan@gmail.com
-
Release v3.74.0 Notes
-
Release v3.74.0 Notes
Co-authored-by: Cursor Agent cursoragent@cursor.com
Co-authored-by: alex-lum alex@cline.bot下载附件