fix(responses): log completed response, not the event, for non-streaming callers

[devin-ai-integration]

TLDR

Problem this solves:

• Non-streaming chat calls through the responses bridge wrote no SpendLogs row
• Every such call logged standard_logging_object not found instead
• Only hits providers whose responses sub-call always returns SSE

How it solves it:

• Responses stream iterator now logs the completed response, not the wrapper event
• Only when the logging object is not in stream mode
• Streaming callers keep getting the event, so their path is unchanged

User Flow

Before: a developer calling a mode: responses model over /v1/chat/completions without streaming gets a good answer that is never billed

1. They send POST https://litellm-domain/v1/chat/completions for a chatgpt/* model, no stream field
2. They get HTTP 200 with the assistant text and a correct usage block
3. They open https://litellm-domain/ui/?page=logs and their request is missing
4. The same request sent with "stream": true does show up, so spend silently depends on the request shape

After: the same non-streaming request is billed like the streaming one

1. They send the same POST https://litellm-domain/v1/chat/completions with no stream field
2. They get the same HTTP 200, text and usage
3. https://litellm-domain/ui/?page=logs now shows the request with its tokens and spend
4. Streaming requests still show up exactly as before, one row each

Relevant issues

Fixes #36426

Linear ticket

Pre-Submission checklist

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review

Screenshots / Proof of Fix

Reproduced against a local proxy backed by postgres, with a mode: responses model whose provider always answers /responses with SSE, which is what forces the bridge to drain a streaming iterator for a non-streaming caller. Not a paid provider call: the failure needs a provider that unconditionally sets stream = True on the responses request, so the backend here is a local SSE server returning a fixed response.completed payload with input_tokens: 11 / output_tokens: 7, and the model is priced in the config so a real cost is computed.

Before, at 457be8f (parent of this branch):

$ curl -s -X POST localhost:4000/v1/chat/completions -H "Authorization: Bearer $KEY" \
    -H 'Content-Type: application/json' \
    -d '{"model":"chatgpt/gpt-5.6-terra","messages":[{"role":"user","content":"baseline"}]}'
{"id":"chatcmpl-...","model":"chatgpt/gpt-5.6-terra","object":"chat.completion","choices":[{"finish_reason":"stop","index":0,"message":{"content":"hello from fake backend","role":"assistant"}}],"usage":{"completion_tokens":7,"prompt_tokens":11,"total_tokens":18}}

$ psql -c 'select count(*) from "LiteLLM_SpendLogs";'   # unchanged, no row for the call above
 count
-------
     3

$ grep -c 'Cost tracking failed' proxy.log
1
# Error in tracking cost callback - Cost tracking failed for model=gpt-5.6-terra.
# Debug info - standard_logging_object not found

After, at ca767b4:

$ curl -s -X POST localhost:4000/v1/chat/completions -H "Authorization: Bearer $KEY" \
    -H 'Content-Type: application/json' \
    -d '{"model":"chatgpt/gpt-5.6-terra","messages":[{"role":"user","content":"final check"}]}'
{"id":"chatcmpl-ff76b496-21ff-44db-8db6-5c3fe25a4688","model":"chatgpt/gpt-5.6-terra","object":"chat.completion","choices":[{"finish_reason":"stop","index":0,"message":{"content":"hello from fake backend","role":"assistant"}}],"usage":{"completion_tokens":7,"prompt_tokens":11,"total_tokens":18}}

$ psql -c 'select call_type,status,spend,prompt_tokens,completion_tokens,"endTime" from "LiteLLM_SpendLogs" order by "endTime" desc limit 1;'
 call_type | status  |  spend  | prompt_tokens | completion_tokens |        endTime
-----------+---------+---------+---------------+-------------------+------------------------
 responses | success | 2.5e-05 |            11 |                 7 | 2026-08-10 19:12:56.77

$ grep -c 'Cost tracking failed' proxy.log
0

Streaming on the same build still bills exactly one row per request, with the same tokens and spend, so the fix does not double count.

Type

🐛 Bug Fix

Caveats (if any)

• Untested against a paid provider that force-streams responses

Link to Devin session: https://app.devin.ai/sessions/5efe36ef464e4140aa82e37d6cf2680d

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

The PR fixes spend logging when a non-streaming caller drains an SSE Responses API iterator by passing the completed response, rather than its wrapper event, to success handlers

• Preserves wrapper events for streaming logging paths
• Adds focused regression coverage for streaming and non-streaming callers
• Ratchets down lint and type-discipline budgets

Confidence Score: 5/5

The PR appears safe to merge, with the non-streaming logging fix isolated from the existing streaming path

The changed branch passes the completed Responses API payload to non-streaming success handlers while preserving the completion event for streaming handlers, and focused tests cover both outcomes

Important Files Changed

Filename	Overview
litellm/responses/streaming_iterator.py	Selectively unwraps completed response events for non-streaming success logging while retaining copied wrapper events for streaming callers
tests/test_litellm/responses/test_streaming_iterator.py	Adds focused asynchronous coverage proving the logged object shape for both stream modes
ruff-strict-budget.json	Ratchets the S110 allowance down by one after removing the broad exception suppression
type-discipline-budget.json	Ratchets the LIT010 allowance down by two for the newly annotated final variables

_{Reviews (1): Last reviewed commit: "fix(responses): log completed response, ..." | Re-trigger Greptile}

[codecov]

Codecov Report

❌ Patch coverage is 93.75000% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/responses/streaming_iterator.py	93.75%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_fix_responses_bridge_spendlog_36426 (6bf2156) with litellm_internal_staging (ade805e)¹}

Footnotes

1. No successful run was found on litellm_internal_staging (c408285) during the generation of this report, so ade805e was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[dalzio]

Could you please close this PR in favor of #40139? The replacement is now conflict-free.

Posted for GP by his AI assistant

[mateo-berri]

Superseded by #41235, which keeps non-streaming callers off the streaming iterator entirely and lands one spend row per call