AI & Agent Dev Bug Sandbox Back to Radar

ChatMistralAI Does Not Populate Standardized Usage_metadata, Causing LangSmith To Ignore Cached_tokens And Overcharge

ChatMistralAI returns token usage only in response_metadata.token_usage, which is deprecated and not parsed by LangSmith for cached token pricing. As a result, LangSmith treats all input tokens as uncached, leading to significant cost overestimation when using prompt caching.

highConfidence 85%Langchain Affected V1.1.5 Fixed V1.1.6

Origin Analysis

The ChatMistralAI integration does not map the Mistral API's usage fields (including prompt_tokens_details.cached_tokens) to LangChain's standardized AIMessage.usage_metadata. LangSmith relies on usage_metadata for cost calculation; the provider-specific response_metadata.token_usage is ignored or deprecated.

1. Install langchain-mistralai 1.1.5 and langsmith 0.7.32.\n2. Set up a Mistral API key and LangSmith environment.\n3. Run the provided script that invokes ChatMistralAI sequentially with prompt_cache_key and compares actual cost (using cached token pricing) against LangSmith's calculated cost.\n4. Observe that LangSmith reports higher input cost because it does not subtract cached tokens; the overcharge percentage is printed.

Fixing Code Block

0x7F 45 4C 46 02 01 01 000x90 12 AB 04 C1 7E 33 08
0xA4 1C 90 3E 77 0B D2 580x11 F6 6A 2C 80 19 DE 44
0x00 00 00 00 03 00 3E 000x01 00 B7 4D 55 90 0C 61
0x6C 69 62 63 2E 73 6F 2E0x36 00 8E 17 42 AA 05 F3
0xDE C0 DE 01 8B 45 FC 890xC7 5D 20 6F 91 03 E4 78
0x13 37 00 4F 9A 2D E1 700x08 B4 5C 22 71 D0 4A 16
0x48 8D 3D 22 01 00 00 E80x91 FF 0A 63 2E 57 B8 04
0x5A 91 C3 0E 44 70 1B 880x02 6F 39 C4 15 80 ED 27
0xFF 15 40 20 60 00 48 890xC3 90 7A 01 4C 28 B1 56
0x0F 1F 44 00 00 55 48 890xE5 C3 18 6D 92 0B 33 A7

Edge Case Audit

Adding usage_metadata may cause double-counting if downstream code also reads response_metadata.token_usage; ensure consumers prefer usage_metadata. The change is backward-compatible but must be tested with both streaming and non-streaming invocations. If rolled back, LangSmith will again ignore cached tokens. No concurrency issues are introduced as this is per-message metadata. Users on older langchain-core versions that do not support usage_metadata will ignore the new field; verify compatibility with langchain-core >= 0.2.