fix(fireworks_ai): restore supports_vision on minimax-m3 in the cost map by devin-ai-integration[bot] · Pull Request #41699 · BerriAI/litellm

devin-ai-integration · 2026-09-17T23:40:35Z

TLDR

Problem this solves:

Price sync chore(prices): sync Together AI prices: 6 models, 6 deprecated [sync failed: Google Gemini] #41570 set fireworks_ai/minimax-m3 to supports_vision: false
Fireworks does serve images on minimax-m3, so proxy users now get a 400
CI on main is red: 11 Fireworks tests still expect vision support

How it solves it:

Sets supports_vision: true on both Fireworks minimax-m3 cost map entries
Restores the minimax-m3 vision pins test(fireworks_ai): stop pinning vision support on minimax-m3 #41627 dropped from the unit tests

User Flow

Before: a developer sending an image to Fireworks minimax-m3 through the proxy gets a 400 saying the model cannot take images, while Fireworks itself answers the same request

They send POST https://litellm-domain/v1/chat/completions with model fireworks-minimax-m3 and one user message holding a text block plus an image_url block
The proxy answers HTTP 400: litellm.BadRequestError: Fireworks AI model accounts/fireworks/models/minimax-m3 does not support image inputs. Use a Fireworks vision model or remove image_url content blocks.
GET https://litellm-domain/model/info lists that deployment with supports_vision: false
The same request sent straight to POST https://api.fireworks.ai/inference/v1/chat/completions returns HTTP 200 and describes the image

After: the same image request goes through the proxy and comes back with the image described

They send POST https://litellm-domain/v1/chat/completions with model fireworks-minimax-m3 and one user message holding a text block plus an image_url block
The proxy answers HTTP 200 with the image described, and usage.prompt_tokens includes the image tokens
GET https://litellm-domain/model/info lists that deployment with supports_vision: true
The same request sent straight to POST https://api.fireworks.ai/inference/v1/chat/completions returns HTTP 200 and describes the image

Relevant issues

Affected release

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

I have added meaningful tests
The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
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 (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Shared setup for every checkpoint below: a fresh worktree at that commit, its own venv, and a proxy booted with --num_workers 2. Every image request goes to Fireworks for real and costs money. The four checkpoints bracket both the regression (the price sync merge) and this fix, so the flag flip is visible from both sides

mat570-proxy-config.yaml:

model_list:
  - model_name: fireworks-minimax-m3
    litellm_params:
      model: fireworks_ai/accounts/fireworks/models/minimax-m3
      api_key: os.environ/FIREWORKS_AI_API_KEY
general_settings:
  master_key: sk-mat570-qa

.env holds only FIREWORKS_AI_API_KEY=<key> and LITELLM_LOCAL_MODEL_COST_MAP=True, so each proxy reads the cost map from its own checkout instead of fetching main's copy at boot

Boot, from the worktree root: set -a && source .env && set +a && python litellm/proxy/proxy_cli.py --config mat570-proxy-config.yaml --port <port> --num_workers 2

chat.json:

{"model":"fireworks-minimax-m3","max_tokens":400,"messages":[{"role":"user","content":[{"type":"text","text":"What animal is in this image? Answer with one word."},{"type":"image_url","image_url":{"url":"https://httpbin.org/image/png"}}]}]}

messages.json:

{"model":"fireworks-minimax-m3","max_tokens":400,"messages":[{"role":"user","content":[{"type":"text","text":"What animal is in this image? Answer with one word."},{"type":"image","source":{"type":"url","url":"https://httpbin.org/image/png"}}]}]}

responses.json:

{"model":"fireworks-minimax-m3","max_output_tokens":400,"input":[{"role":"user","content":[{"type":"input_text","text":"What animal is in this image? Answer with one word."},{"type":"input_image","image_url":"https://httpbin.org/image/png"}]}]}

Checkpoint table (the Before and After blocks below are the merge base and the tip rows in full):

Checkpoint	Commit	Expected	POST /v1/chat/completions	POST /v1/messages	POST /v1/responses	GET /model/info
main before #41570 merged	`decbb96382`	200 on every image request, `supports_vision: true`	PASS 200, `Pig`, prompt_tokens 155	PASS 200, `Pig`, input_tokens 1	PASS 200, `Pig`, input_tokens 155	PASS 200, `supports_vision: true`
#41570 merge commit (the price sync)	`87650bfe67`	400 on chat and messages, `supports_vision: false`; responses never hit the gate	FAIL 400, `does not support image inputs`	FAIL 400, `does not support image inputs`	PASS 200, `Pig`, input_tokens 155	FAIL 200, `supports_vision: false`
this PR's merge base on main	`e5aa10a1ea`	400 on chat and messages, `supports_vision: false`; responses never hit the gate	FAIL 400, `does not support image inputs`	FAIL 400, `does not support image inputs`	PASS 200, `Pig`, input_tokens 155	FAIL 200, `supports_vision: false`
this PR's tip	`99b83a2d52`	200 on every image request, `supports_vision: true`	PASS 200, `Pig`, prompt_tokens 155	PASS 200, `Pig`, input_tokens 155	PASS 200, `Pig`, input_tokens 155	PASS 200, `supports_vision: true`