SaaS Metrics
Gemini Executive Synthesis
Connectivity and model compatibility issues with MCP Codex and various GPT models
Technical Positioning
Flexible, multi-LLM agent platform for autonomous ML research
SaaS Insight & Market Implications
This issue exposes critical interoperability and compatibility failures within ARIS's multi-LLM agent framework. Users are encountering 400 errors due to unsupported model configurations (e.g., `gpt-5.4-xhigh`, `gpt-4o` with Codex via a ChatGPT account). The system's fallback mechanism is failing, leading to an inability to determine which model is ultimately used. This directly undermines ARIS's core value proposition of "cross-model review loops" and "no lock-in," as model flexibility is compromised by integration complexities. Such errors create significant operational friction and erode trust in the platform's ability to reliably leverage diverse LLM agents.
Proprietary Technical Taxonomy
Raw Developer Origin & Technical Request
GitHub Issue
Mar 24, 2026
Repo: wanshuiyin/Auto-claude-code-research-in-sleep
mcp codex连不上的情况
有时会出现codex返回400错误代码的情况，然后一路从5.4降到4o，最后不显示模型了，那最后用的什么模型呢？
> Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.4-xhigh' model is not supported when using Codex with
> a ChatGPT account."}}
> Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-4o' model is not supported when using Codex with a
> ChatGPT account."}}
因为tz的问题吗？有大佬遇到过类似情况么。
Developer Debate & Comments
> 有时会出现codex返回400错误代码的情况，然后一路从5.4降到4o，最后不显示模型了，那最后用的什么模型呢？ > > > Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.4-xhigh' model is not supported when using Codex with > > a ChatGPT account."}} > > > Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-4o' model is not supported when using Codex with a > > ChatGPT account."}} > > 因为tz的问题吗？有大佬遇到过类似情况么。 请问开了GPT plus 嘛～ 我今天晚上用还是好的诶
> > 有时会出现codex返回400错误代码的情况，然后一路从5.4降到4o，最后不显示模型了，那最后用的什么模型呢？ > > > Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.4-xhigh' model is not supported when using Codex with > > > a ChatGPT account."}} > > > > > > > Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-4o' model is not supported when using Codex with a > > > ChatGPT account."}} > > > > > > 因为tz的问题吗？有大佬遇到过类似情况么。 > > 请问开了GPT plus 嘛～ 我今天晚上用还是好的诶 开了的大佬，网页版能正常用。直接问了claude，破案了😂 > 配置确实是 model = "gpt-5.4"。 > > 问题出在我第一次调用时手动指定了 model: "gpt-5.4-xhigh"（带 -xhigh 后缀），这不是一个合法的模型ID，所以报错。后续又试了 o3、gpt-4o，同样不被 ChatGPT > 账户支持。 > > 最后成功的调用没有传 model 参数，于是 Codex 用了配置文件里的默认值 gpt-5.4，这才走通了。 > > 总结：你的配置没问题，是我的错——不该手动覆盖 model 参数，应该让它走默认配置。后续调用我会直接用默认的 gpt-5.4。
> > > 有时会出现codex返回400错误代码的情况，然后一路从5.4降到4o，最后不显示模型了，那最后用的什么模型呢？ > > > > Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.4-xhigh' model is not supported when using Codex with > > > > a ChatGPT account."}} > > > > > > > > > > Error: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-4o' model is not supported when using Codex with a > > > > ChatGPT account."}} > > > > > > > > > 因为tz的问题吗？有大佬遇到过类似情况么。 > > > > > > 请问开了GPT plus 嘛～ 我今天晚上用还是好的诶 > > 开了的大佬，网页版能正常用。直接问了claude，破案了😂 > > > 配置确实是 model = "gpt-5.4"。 > > 问题出在我第一次调用时手动指定了 model: "gpt-5.4-xhigh"（带 -xhigh 后缀），这不是一个合法的模型ID，所以报错。后续又试了 o3、gpt-4o，同样不被 ChatGPT > > 账户支持。 > > 最后成功的调用没有传 model 参数，于是 Codex 用了配置文件里的默认值 gpt-5.4，这才走通了。 > > 总结：你的配置没问题，是我的错——不该手动覆盖 model 参数，应该让它走默认配置。后续调用我会直接用默认的 gpt-5.4。 你的这个model = "gpt-5.4"，是在哪里改的，我这个自动跳gpt5.2
这个 400 错误是 Codex 账户类型限制导致的，不是 ARIS 的问题： > The 'gpt-5.4-xhigh' model is not supported when using Codex with a ChatGPT account. **原因**：ChatGPT Plus 订阅的 Codex 对可用模型有限制，部分高端模型（gpt-5.4）可能不在支持列表中。 **解决方案**： 1. **检查 Codex 支持的模型列表**：运行 `codex` 看可用模型 2. **如果 gpt-5.4 不可用**，可以在 skill 中手动指定较低的模型（如 `o3`） 3. **或者用 llm-chat MCP 替代**：通过 OpenAI API key 直连（不经 Codex），支持所有模型。详见 `mcp-servers/llm-chat/`，同时可用 `tools/convert_skills_to_llm_chat.py` 自动转换所有 skill 如果还有问题欢迎重新开 issue。
Adjacent Repository Pain Points
Other highly discussed features and pain points extracted from wanshuiyin/Auto-claude-code-research-in-sleep.
Extracted Positioning
ARIS compatibility with OpenAI Codex.
Maintaining broad LLM agent compatibility ('works with Claude Code, Codex, OpenClaw, or any LLM agent') to offer flexibility and avoid vendor lock-in.
Top Replies
> No description provided. 我们即将给一个md 文档描述怎么适配，但是现在git仓库有点问题，不能fork 请稍等
> No description provided. 现在已经适配Cursor 和单独一个Codex Subagent review，欢迎体验～
> > No description provided. > > 现在已经适配Cursor 和单独一个Codex Subagent review，欢迎体验～ 出现代码问题时，能否让他把问题写进code_guide.md文件，然后等待人接入，如果一段时间没人介入就自动cli端...
Extracted Positioning
ARIS integration with Feishu (飞书) via Claude Code in bidirectional interactive mode.
Enabling seamless, bidirectional communication and interaction between ARIS (using Claude Code) and enterprise collaboration platforms like Feishu, supporting 'autonomous ML research' within existing workflows.
Top Replies
感谢反馈！飞书这个东西我昨天也配了很久跑通 （所以飞书我最建议是false或者只推送进度，有问题回工作做hhh）。 不过这个问题我们在搭建时也遇到过，能收到消息但不回复，我让cc总结了一下， 希望对你有用，或者...
**桥接输出如下图，没有显示received message之类的日志， 同时也没有/path/to/feishu-claude-code/logs/*.log这一路径**
Extracted Positioning
ARIS research pipeline automation with GLM-5 + MiniMAX 2.5 LLM combination.
Achieving full, uninterrupted automation for research pipelines, as implied by 'AUTO_PROCEED: true' and the 'autonomous' nature of ARIS.
Top Replies
我opus 4.6 + gpt 5.4 xhigh fast /research-pipeline — AUTO_PROCEED: true 也会断
好奇，这个项目真的可以拿下a会吗？
Extracted Positioning
ARIS (Auto-Research-In-Sleep) with 阿里百炼 (Ali Bailian) LLM agent.
Ensuring stable, uninterrupted execution of long-running autonomous ML research tasks, particularly when integrating with specific LLM providers and network configurations (proxies, SSH).
Top Replies
同样的问题，执行到一半就停了
一样的情况，不稳定
Extracted Positioning
Workflow 3 usage for paper writing on Windows
Accessible autonomous ML research and content generation across platforms
Frequently Asked Questions
Market intelligence mapped to Connectivity and model compatibility issues with MCP Codex and various GPT models.
What problem does Connectivity and model compatibility issues with MCP Codex and various GPT models solve?
Based on our AI analysis of the original developer request, its primary technical positioning is: Flexible, multi-LLM agent platform for autonomous ML research
How is the developer community reacting to Connectivity and model compatibility issues with MCP Codex and various GPT models?
Yes, we have tracked 3 direct responses and active debates regarding this specific topic originating from GitHub Issue.
What architecture is tied to Connectivity and model compatibility issues with MCP Codex and various GPT models?
Our proprietary extraction maps Connectivity and model compatibility issues with MCP Codex and various GPT models to adjacent architectural concepts including mcp codex, 400错误代码, gpt-5.4-xhigh, gpt-4o.
Engagement Signals
Cross-Market Term Frequency
Quantifies the cross-market adoption of foundational terms like invalid_request_error and gpt-4o by tracking occurrence frequency across active SaaS architectures and enterprise developer debates.