fix(llm): correct gpt-4o-mini context window 200000 -> 128000

[Shxiao101]

Related issue

Fixes #7293

Summary

gpt-4o-mini's official context window is 128,000 tokens (OpenAI announcement and API docs); litellm's own model database also reports max_input_tokens = 128000. The shared LLM_CONTEXT_WINDOW_SIZES table and the OpenAI/Azure provider-local tables list 200000 — apparently copied from the neighboring o3-mini / o4-mini entries (both 200000). The same tables list gpt-4o as 128000, so the smaller mini having a larger window is internally inconsistent.

Because get_context_window_size() applies CONTEXT_WINDOW_USAGE_RATIO = 0.85, crews using gpt-4o-mini resolved their usable window to 170000 instead of 108800 on all three provider paths (the litellm-path lookup in llm.py iterates without break, so gpt-4o-mini resolves to its own entry), letting conversation history grow past the model's real 128k limit and failing with API 400s on long-running crews.

This changes only the gpt-4o-mini value in three places — no key additions, removals, or reordering, so lookups for any other model are unaffected.

Note for review: open PR #6603 (longest-prefix matching) adds a test asserting gpt-4o-mini == 200000; if it merges first, its assertion needs the same correction.

Verification

• Tests added or updated for the changed behavior → covered by the existing context-window tests (pytest lib/crewai/tests/test_llm.py -k context_window, 10 passed); happy to add a dedicated gpt-4o-mini regression test if preferred
• Relevant tests and quality checks pass locally

Local checks:

import os
os.environ.setdefault("OPENAI_API_KEY", "test-key")
from crewai.llm import LLM

LLM(model="gpt-4o-mini", is_litellm=True).get_context_window_size()  # 170000 -> 108800
LLM(model="openai/gpt-4o-mini").get_context_window_size()            # 170000 -> 108800

• uv run ruff check <changed files> and uv run ruff format --check <changed files>: clean
• pytest lib/crewai/tests/test_llm.py -q: 89 passed, 3 skipped (2 pre-existing environment errors on Windows, reproduced on unmodified main)
• Full-repo scan confirms the three changed entries are the only gpt-4o-mini window-size occurrences

Additional context

• No existing issue tracked this value; [BUG] Gemini gemini-2.0-flash-thinking returns the wrong context window (891289 instead of 27852) #7129 is the same class of bug for a different model, which motivated opening [BUG] gpt-4o-mini context window is set to 200000 instead of the official 128000 #7293 first.
• Disclosure per CONTRIBUTING: this PR was prepared with AI assistance and human-reviewed; please apply the llm-generated label (contributors without triage permissions cannot add labels themselves).

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 67e22152-647b-4ea7-9696-7ebcf8b4917e

📥 Commits

Reviewing files that changed from the base of the PR and between 143e902 and 9478f42.

📒 Files selected for processing (3)

• lib/crewai/src/crewai/llm.py
• lib/crewai/src/crewai/llms/providers/azure/completion.py
• lib/crewai/src/crewai/llms/providers/openai/completion.py

Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review.

---

📝 Walkthrough

Walkthrough

The change corrects the gpt-4o-mini context-window value from 200000 to 128000 in the shared LLM mapping and the OpenAI and Azure provider mappings.

Changes

Context window correction

Layer / File(s)	Summary
Update context-window mappings lib/crewai/src/crewai/llm.py, lib/crewai/src/crewai/llms/providers/openai/completion.py, lib/crewai/src/crewai/llms/providers/azure/completion.py	The gpt-4o-mini context-window value is changed from 200000 to 128000 in all three mappings. The OpenAI provider applies the usage ratio to the corrected value.

Suggested reviewers: lorenzejay, joaomdmoura

Merge Risk: ⚪ Minimal · up to 9478f

This corrects the gpt-4o-mini context limit across shared, OpenAI, and Azure paths, preventing oversized histories from exceeding the model limit. The change is ready to merge.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely identifies the correction to the gpt-4o-mini context window.
Description check	✅ Passed	The description includes the related issue, problem rationale, affected paths, verification results, test status, and additional context.
Linked Issues check	✅ Passed	The changes satisfy issue #7293 by updating gpt-4o-mini from 200000 to 128000 in the shared, OpenAI, and Azure tables. The expected resolved window is 108800.
Out of Scope Changes check	✅ Passed	All changes are limited to the three context-window entries required by issue #7293. No unrelated code changes are reported.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 3 files.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Vidit-Ostwal]

Approved. Official OpenAI docs list gpt-4o-mini at 128,000 context tokens (and 16,384 max output). The previous 200000 matches Tier 1 TPM, not the context window. The three table updates (LiteLLM, native OpenAI, Azure) are the right places and leave other model lookups alone.

[Vidit-Ostwal]

Thanks for things, cheers