litellm - 💡(How to fix) Fix [Bug]: /v1/responses endpoint returns x-litellm-response-cost: 0 while spend_logs records correct cost [1 participants]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

The /v1/responses (OpenAI Responses API) endpoint returns x-litellm-response-cost: 0 in the response header, while the /v1/chat/completions endpoint returns the correct cost for the same model. Interestingly, the spend is correctly recorded in LiteLLM_SpendLogs — it's only the response header that's missing the cost.

This is similar to #23981 (which reports the same issue for /v1/messages), but affects the OpenAI Responses API path.

Environment

• LiteLLM version: v1.82.3-stable (also reproduced on v1.83.7)
• Edition: Community (Docker, ghcr.io/berriai/litellm:main-stable)
• Deployment: Self-hosted Docker
• Backend: OpenAI-compatible provider (Codex gateway)
• Database: PostgreSQL

Reproduction

1. Call /v1/chat/completions (works correctly)

curl -s -D- http://litellm:4000/v1/chat/completions \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4","messages":[{"role":"user","content":"Say hello"}]}'

Response headers (correct):

x-litellm-response-cost: 0.0011335    ✅ present & correct

2. Call /v1/responses (broken)

curl -s -D- http://litellm:4000/v1/responses \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.4","input":"Say hello"}'

Response headers (broken):

x-litellm-response-cost: 0            ❌ always 0

3. Spend logs DO record the correct cost

Querying /spend/logs confirms the cost was correctly computed and stored:

{
  "request_id": "resp_089ebd3558db3572...",
  "spend": 0.0399475,
  "model": "gpt-5.4"
}

Expected behavior

x-litellm-response-cost should return the actual computed cost for /v1/responses, just as it does for /v1/chat/completions.

Root cause hypothesis

The Responses API appears to go through a pass-through / different code path that doesn't call the same cost computation logic used by the Chat Completions handler before constructing the response headers. The cost IS computed asynchronously (hence it appears in spend_logs), but is not available at header-write time.

Additional note: request_id format mismatch

A secondary issue: for /v1/responses, the request_id stored in LiteLLM_SpendLogs uses a hex-encoded short format of the provider's response_id (e.g., resp_089ebd3558db3572...), while the actual provider response contains a longer Base64-like ID (e.g., resp_0Ci7hJsa-_bXKDXne0WxR...). This makes it impossible for downstream applications to look up spend logs by the provider's original response_id. For Chat Completions, the request_id is a standard UUID that's also returned in the x-litellm-call-id header, making lookups straightforward.

Relevant log/trace info

• Tested on both v1.82.3-stable and v1.83.7 — same behavior
• Non-streaming mode
• store_prompts_in_spend_logs: true enabled in config

Are you a LiteLLM Enterprise user?

No

Twitter / LinkedIn details

No response