OpenAI newer models fail because requests use max_tokens instead of max_completion_tokens

[xwbcl123]

Summary

The OpenAI-compatible provider fails its connection test with newer OpenAI models because the request body sends the deprecated max_tokens parameter.

OpenAI returns:

API error (400): {
  "error": {
    "message": "Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.",
    "type": "invalid_request_error",
    "param": "max_tokens",
    "code": "unsupported_parameter"
  }
}

Environment

• Easy-Translater extension version: 1.0.0
• Upstream commit tested: 8d0378258c255dcde8c479e7da0d91aee9cefa3f
• Provider: OpenAI compatible
• API base URL: https://api.openai.com/v1/
• Model: gpt-5.4-mini
• Browser: Google Chrome on macOS

Steps to reproduce

1. Open the extension settings.
2. Select OpenAI compatible as the API provider.
3. Set the API URL to https://api.openai.com/v1/.
4. Enter a valid OpenAI API key.
5. Set the model to gpt-5.4-mini.
6. Run the connection test.

Actual behavior

The connection test fails with HTTP 400 and reports that max_tokens is unsupported.

Expected behavior

The connection test and translations should succeed with newer OpenAI models that require max_completion_tokens.

Root cause

lib/translator.js hardcodes max_tokens: 4096 in the OpenAI request body (around line 84). The same parameter is also hardcoded in other OpenAI-compatible request paths (around lines 103 and 152).

OpenAI's current Chat Completions API reference marks max_tokens as deprecated in favor of max_completion_tokens, and notes that parameter support differs by model:

https://developers.openai.com/api/reference/resources/chat/subresources/completions/methods/create

Suggested fix

Please consider:

1. Using max_completion_tokens for native OpenAI requests and models that require it.
2. Preserving compatibility with third-party OpenAI-compatible endpoints that may still expect max_tokens, for example through provider/model-aware parameter selection or a targeted retry when the API returns unsupported_parameter.
3. Centralizing construction of the token-limit parameter instead of hardcoding it in multiple request paths.
4. Adding a regression test covering both parameter variants.

No API key or other credential is included in this report.

[xwbcl123]

补充一份中文说明，方便作者快速定位问题：

我在 Chrome 中试用 Easy-Translater 1.0.0，配置如下：

• API 提供商：OpenAI 兼容
• API 地址：https://api.openai.com/v1/
• 模型：gpt-5.4-mini

点击连接测试后返回 HTTP 400：

Unsupported parameter: 'max_tokens' is not supported with this model.
Use 'max_completion_tokens' instead.

检查代码后发现，lib/translator.js 的 OpenAI 请求体中固定使用了：

max_tokens: 4096

目前 OpenAI 的较新模型要求使用 max_completion_tokens。建议：

1. 对 OpenAI 官方接口及要求新参数的模型使用 max_completion_tokens；
2. 考虑到其他 OpenAI-compatible 服务可能仍使用 max_tokens，可以根据 provider/model 选择参数，或在收到 unsupported_parameter 时进行一次针对性重试；
3. 将 token 上限参数的构建逻辑集中到一个函数中，避免当前多处硬编码；
4. 增加覆盖新旧参数的回归测试。

感谢开发和维护这个插件。如需要，我也可以进一步协助验证修复版本。

注：报告中没有包含任何 API Key 或其他凭据。