fix: expose reasoning effort fields in get_model_info + add together_ai/gpt-oss-120b by avarga1 · Pull Request #25263 · BerriAI/litellm

avarga1 · 2026-04-07T03:31:21Z

Changes

Fix 1 — `get_model_info` drops `supports_xhigh_reasoning_effort` / `supports_none_reasoning_effort` (closes #25096)

_get_model_info_helper in litellm/utils.py constructs a ModelInfoBase by explicitly mapping fields from _model_info. supports_none_reasoning_effort and supports_xhigh_reasoning_effort were present in ProviderSpecificModelInfo and in the JSON data (e.g. gpt-5.4 has both set to true) but were never passed through in the constructor call — so get_model_info() always returned None for them.

Fix: add the two missing .get() calls alongside supports_reasoning.

# before
supports_reasoning=_model_info.get("supports_reasoning", None),
supports_computer_use=_model_info.get("supports_computer_use", None),

# after
supports_reasoning=_model_info.get("supports_reasoning", None),
supports_none_reasoning_effort=_model_info.get("supports_none_reasoning_effort", None),
supports_xhigh_reasoning_effort=_model_info.get("supports_xhigh_reasoning_effort", None),
supports_computer_use=_model_info.get("supports_computer_use", None),

Fix 2 — `together_ai/openai/gpt-oss-120b` missing from model registry (closes #25132)

together_ai/openai/gpt-oss-120b had no entry in model_prices_and_context_window.json, so LiteLLM treated it as not supporting reasoning_effort and raised UnsupportedParamsError. Added the entry with "supports_reasoning": true, modelled after the existing together_ai/openai/gpt-oss-20b and azure_ai/gpt-oss-120b entries. Pricing sourced from https://www.together.ai/models/gpt-oss-120b.

Test plan

litellm.get_model_info("gpt-5.4")["supports_xhigh_reasoning_effort"] returns True
litellm.get_model_info("gpt-5.4")["supports_none_reasoning_effort"] returns True
litellm.completion(model="together_ai/openai/gpt-oss-120b", messages=[...], reasoning_effort="low") no longer raises UnsupportedParamsError

🤖 Generated with Claude Code

…r_ai/gpt-oss-120b - litellm/utils.py: pass supports_none_reasoning_effort and supports_xhigh_reasoning_effort through _get_model_info_helper so get_model_info() returns them (previously silently dropped). Fixes BerriAI#25096. - model_prices_and_context_window.json: add together_ai/openai/gpt-oss-120b with supports_reasoning: true so reasoning_effort is accepted for this model without requiring drop_params. Fixes BerriAI#25132. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

vercel · 2026-04-07T03:31:26Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Apr 8, 2026 2:13pm

CLAassistant · 2026-04-07T03:31:30Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ avarga1
❌ Austin Varga

Austin Varga seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

codspeed-hq · 2026-04-07T03:33:50Z

Merging this PR will not alter performance

✅ 16 untouched benchmarks

_{Comparing avarga1:fix/model-info-reasoning-fields-and-gpt-oss-120b (374eabb) with main (62757ff)}

codecov · 2026-04-07T03:34:31Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

avarga1 · 2026-04-07T03:35:29Z

Note on the format check failure: this is a pre-existing repo-wide issue tracked in #25181 — 328 files fail the formatter on main before any of our changes. Our diff is only 2 lines in utils.py and 13 lines in the JSON file. Happy to reformat our specific changes if that's helpful, but the CI failure isn't caused by this PR.

greptile-apps · 2026-04-07T03:35:36Z

Greptile Summary

This PR fixes two related issues: (1) get_model_info() was silently dropping supports_none_reasoning_effort and supports_xhigh_reasoning_effort because _get_model_info_helper never included them in the ModelInfoBase constructor call; and (2) together_ai/openai/gpt-oss-120b was missing supports_reasoning: true in the model registry, causing UnsupportedParamsError when reasoning_effort was passed.

litellm/utils.py: Two .get() calls added to _get_model_info_helper to pass the missing reasoning-effort fields through to ModelInfoBase, alongside the existing supports_reasoning field.
model_prices_and_context_window.json: Consolidated the together_ai/openai/gpt-oss-120b entry (previously a duplicate) into a single, correct record with supports_reasoning: true, max_output_tokens: 131072, and a source URL.
litellm/model_prices_and_context_window_backup.json: Synced with the same corrected entry so the fallback code path is also fixed.

Previous review concerns (duplicate JSON key and out-of-sync backup file) have been fully addressed by the subsequent commits in this PR. The fix is minimal and correct.

Confidence Score: 5/5

Safe to merge — all three files are correct, previous duplicate-key and backup-sync issues are resolved, and the utils.py passthrough fix is minimal and accurate.

All findings are P2 (missing regression tests). No logic errors, no security issues, and the previously flagged duplicate key and unsynced backup file have both been addressed in subsequent commits.

No files require special attention; the only suggestion is adding a unit test to tests/test_litellm/test_utils.py.

Vulnerabilities

No security concerns identified. Changes are limited to JSON model registry data and a read-only model info lookup helper; no auth, input validation, or data-handling paths are affected.

Important Files Changed

Filename	Overview
litellm/utils.py	Adds two missing `.get()` calls in `_get_model_info_helper` to expose `supports_none_reasoning_effort` and `supports_xhigh_reasoning_effort` from the JSON registry — correct and minimal fix.
model_prices_and_context_window.json	Consolidated previously-duplicate `together_ai/openai/gpt-oss-120b` entry into a single correct record; now has `supports_reasoning: true`, `max_output_tokens: 131072`, and source URL.
litellm/model_prices_and_context_window_backup.json	Backup file synced with the corrected `together_ai/openai/gpt-oss-120b` entry; fallback code path will now also correctly expose `supports_reasoning: true`.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["litellm.get_model_info(model)"] --> B["_get_model_info_helper()"]
    B --> C["Lookup in model_prices_and_context_window.json\n(fallback: backup.json)"]
    C --> D["_model_info dict\n(includes supports_reasoning,\nsupports_none_reasoning_effort,\nsupports_xhigh_reasoning_effort)"]
    D --> E["ModelInfoBase constructor"]
    E -->|"Before fix: missing fields"| F["supports_none_reasoning_effort = None\nsupports_xhigh_reasoning_effort = None"]
    E -->|"After fix: fields passed through"| G["supports_none_reasoning_effort = True/False\nsupports_xhigh_reasoning_effort = True/False"]
    G --> H["Returned to caller"]
    F --> I["Bug: caller always sees None"]

_{Reviews (3): Last reviewed commit: "fix: link commit to GitHub account for C..." | Re-trigger Greptile}

greptile-apps · 2026-04-07T03:35:40Z

+    "together_ai/openai/gpt-oss-120b": {
+        "input_cost_per_token": 1.5e-07,
+        "litellm_provider": "together_ai",
+        "max_input_tokens": 131072,
+        "max_output_tokens": 131072,
+        "max_tokens": 131072,
+        "mode": "chat",
+        "output_cost_per_token": 6e-07,
+        "source": "https://www.together.ai/models/gpt-oss-120b",
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },


Duplicate key — existing entry not removed

The key "together_ai/openai/gpt-oss-120b" already exists earlier in this file (~line 28496, present before this PR), with max_input_tokens: 128000 and no supports_reasoning field. This PR appends a second entry instead of updating the existing one.

JSON does not permit duplicate keys (RFC 8259 §4 leaves the behavior undefined). Python's json.loads silently picks the last occurrence — so the fix works at runtime — but:

The orphaned first entry has a conflicting max_input_tokens value (128000 vs 131072).

Strict JSON parsers, linters, and jq will flag or reject the file.

It will confuse future maintainers who see two entries for the same model.

Fix: Remove (or update in place) the pre-existing entry at ~line 28496 so the key appears exactly once with the correct values.

greptile-apps · 2026-04-07T03:35:41Z

+    "together_ai/openai/gpt-oss-120b": {
+        "input_cost_per_token": 1.5e-07,
+        "litellm_provider": "together_ai",
+        "max_input_tokens": 131072,
+        "max_output_tokens": 131072,
+        "max_tokens": 131072,
+        "mode": "chat",
+        "output_cost_per_token": 6e-07,
+        "source": "https://www.together.ai/models/gpt-oss-120b",
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },


Backup file not updated

litellm/model_prices_and_context_window_backup.json is the fallback used when the primary file cannot be loaded. It still has only the old entry for "together_ai/openai/gpt-oss-120b" — max_input_tokens: 128000, no supports_reasoning: true, no max_output_tokens/max_tokens. Any code path that falls back to the backup will still trigger the UnsupportedParamsError this PR is fixing.

Please add the corrected entry to litellm/model_prices_and_context_window_backup.json as well.

…sync backup file

vercel bot deployed to Preview April 7, 2026 03:32 View deployment

greptile-apps bot reviewed Apr 7, 2026

View reviewed changes

fix: consolidate duplicate together_ai/openai/gpt-oss-120b entry and …

9169f8c

…sync backup file

vercel bot deployed to Preview April 7, 2026 18:50 View deployment

fix: link commit to GitHub account for CLA verification

374eabb

vercel bot deployed to Preview April 8, 2026 14:13 View deployment

krrish-berri-2 changed the base branch from main to litellm_oss_staging_04_08_2026 April 9, 2026 04:33

krrish-berri-2 merged commit 541e81d into BerriAI:litellm_oss_staging_04_08_2026 Apr 9, 2026
49 of 51 checks passed

jeanibarz mentioned this pull request Apr 10, 2026

fix(together_ai): support reasoning_effort for gpt-oss models #25520

Open

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: expose reasoning effort fields in get_model_info + add together_ai/gpt-oss-120b#25263

fix: expose reasoning effort fields in get_model_info + add together_ai/gpt-oss-120b#25263
krrish-berri-2 merged 3 commits intoBerriAI:litellm_oss_staging_04_08_2026from
avarga1:fix/model-info-reasoning-fields-and-gpt-oss-120b

avarga1 commented Apr 7, 2026

Uh oh!

vercel bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

CLAassistant commented Apr 7, 2026 •

edited

Loading

Uh oh!

codspeed-hq bot commented Apr 7, 2026 •

edited

Loading

Uh oh!

codecov bot commented Apr 7, 2026

Uh oh!

avarga1 commented Apr 7, 2026

Uh oh!

greptile-apps bot commented Apr 7, 2026 •

edited

Loading

Vulnerabilities

Important Files Changed

Uh oh!

greptile-apps bot Apr 7, 2026

Uh oh!

greptile-apps bot Apr 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

avarga1 commented Apr 7, 2026

Changes

Fix 1 — get_model_info drops supports_xhigh_reasoning_effort / supports_none_reasoning_effort (closes #25096)

Fix 2 — together_ai/openai/gpt-oss-120b missing from model registry (closes #25132)

Test plan

Uh oh!

vercel bot commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

CLAassistant commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codspeed-hq bot commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Uh oh!

codecov bot commented Apr 7, 2026

Codecov Report

Uh oh!

avarga1 commented Apr 7, 2026

Uh oh!

greptile-apps bot commented Apr 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Vulnerabilities

Important Files Changed

Flowchart

Uh oh!

greptile-apps bot Apr 7, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Apr 7, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Fix 1 — `get_model_info` drops `supports_xhigh_reasoning_effort` / `supports_none_reasoning_effort` (closes #25096)

Fix 2 — `together_ai/openai/gpt-oss-120b` missing from model registry (closes #25132)

vercel bot commented Apr 7, 2026 •

edited

Loading

CLAassistant commented Apr 7, 2026 •

edited

Loading

codspeed-hq bot commented Apr 7, 2026 •

edited

Loading

greptile-apps bot commented Apr 7, 2026 •

edited

Loading