Add support for "max_completion_tokens" for OpenAI endpoint

[fahim-choudhury]

OpenCodeReview Version

open-code-review v1.1.12 (fe647bc) linux/amd64 built at: 2026-06-03T01:35:05Z https://github.com/alibaba/open-code-review

Operating System

Linux (x86_64)

Installation Method

npm (global)

LLM Provider

OpenAI

Bug Description

When I run ocr llm test, the following error is observed:

Error: llm request failed: API error 400: Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.

Steps to Reproduce

Here's the configuration in ~/.zshrc

# Open Code Review
export OCR_LLM_URL=https://api.openai.com/v1/chat/completions
export OCR_LLM_TOKEN=$OPENAI_API_KEY
export OCR_LLM_MODEL=gpt-5.4
export OCR_USE_ANTHROPIC=false

Expected Behavior

The tool should support max_completion_tokens and ocr llm test should succeed.

Logs / Error Output

Additional Context

No response

[stay-foolish-forever]

Thanks for the detailed report, @fahim-choudhury — and apologies for the trouble.
The root cause here is that our current implementation calls the model APIs directly from our own code, rather than wrapping the official provider client SDKs. As newer models from OpenAI (and other providers) introduce parameter changes — such as replacing max_tokens with max_completion_tokens for the latest model families — our hand-rolled request layer fails to keep up, and compatibility issues like this one surface across different models and providers.
We're treating this as our highest-priority issue. The fix is to migrate our LLM request layer onto the official client libraries so we automatically inherit upstream parameter and protocol changes, which will resolve this issue along with similar compatibility gaps across providers. We'll ship it as soon as possible and update this thread once a release is available.
Thanks again for flagging it.

[fahim-choudhury]

@stay-foolish-forever Thank you for the prompt response. Looking forward to the fix! 🤞🏽

[lizhengfeng101]

Thanks for reporting this, @fahim-choudhury! This has been fixed in v1.2.3 — we now use max_completion_tokens instead of the deprecated max_tokens, so it works with newer OpenAI models (gpt-5.x, o-series, etc.).

Please update to the latest version to resolve the issue. Feel free to reopen if you run into any further problems!

[fahim-choudhury]

Thank you, @stay-foolish-forever and @lizhengfeng101, for the prompt fix, it's working as expected in v1.2.3 🎉