feat(sambanova): add MiniMax-M3, update model limits, support extra chat params

[jhpiedrahitao]

TLDR

Problem this solves:

• sambanova/MiniMax-M3 is missing from the model map, so it has no cost or limits
• Several SambaNova chat params (seed, n, penalties, reasoning_effort) get rejected or dropped
• Some max output limits are wrong (DeepSeek-V3.1/V3.2, Llama-3.3-70B)
• The X-Integration-Source header is not sent, so SambaNova can't tell LiteLLM traffic apart
• Retired models (DeepSeek-R1, Llama-4-Maverick) are still listed

How it solves it:

• Adds sambanova/MiniMax-M3 (1M context, tools, reasoning, response schema)
• Fixes max output tokens to match SambaNova's live /v1/models
• Removes the two retired models from both model JSON files
• Allows the extra OpenAI-compatible params; reasoning_effort only on reasoning models
• Sends X-Integration-Source (default litellm), overridable per request or by env var

Intentional product change: sambanova/DeepSeek-R1 and sambanova/Llama-4-Maverick-17B-128E-Instruct are no longer in the model map, so cost and limit lookups for them stop working.

User Flow

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/unit/<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
• X ] 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

Setup: SAMBANOVA_API_KEY set; proxy started with
python litellm/proxy/proxy_cli.py --config litellm/proxy/dev_config.yaml --detailed_debug
and a sambanova/MiniMax-M3 entry in the config.

Before

MiniMax-M3 chat completion with seed

1. curl http://localhost:4000/v1/chat/completions -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" -d '{"model":"sambanova/MiniMax-M3","messages":[{"role":"user","content":"hi"}],"seed":7}'
2. Observed output: <paste output; show missing cost / dropped seed>

Integration header

1. Same call with --detailed_debug, inspect the outgoing request headers
2. Observed output: <paste; no X-Integration-Source>

After

MiniMax-M3 chat completion with seed

1. Same curl as Before
2. Observed output: <paste; 200 and non-zero cost in response headers / UI logs>

Integration header

1. Same call with --detailed_debug, inspect the outgoing request headers
2. Observed output: <paste; X-Integration-Source: litellm>
3. Repeat with SAMBANOVA_INTEGRATION_SOURCE=my-framework set
4. Observed output: <paste; X-Integration-Source: my-framework>

Type

🆕 New Feature

Caveats (if any)

QA runbook

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[greptile-apps]

Confidence Score: 4/5

[Medium risk] Updates SambaNova model catalog and chat parameter support.

The PR is not ready to merge until the outstanding repository requirement for tests of external facts is satisfied

Findings

1.  Tests pin outside facts ▶

Summary

The PR adds SambaNova MiniMax-M3 metadata, updates model limits and supported chat parameters, removes two retired model entries, and changes integration-source header selection

• The latest changes address the earlier body-override and header-casing findings and remove the flagged test comment
• The existing model-metadata tests still pin vendor facts without dated sources

_{Reviews (2) · Last reviewed commit: "fix(sambanova): handle integration sourc..."}

[greptile-apps]

Tests pin outside facts These assertions hardcode vendor prices and limits; others require retired models to remain absent. Repository rules require owned invariants or dated sources before merging

Context Used: AGENTS.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing jhpiedrahitao:feat/sambanova-update-2026-10 (fa6090b) with main (e2971e0)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[jhpiedrahitao]

docs PR BerriAI/litellm-docs#2092

[snova-jorgep]

@greptileai