[Bug] Custom OpenAI-compatible provider errors with "Unknown parameter: 'reasoningSummary'" for GPT-5 series models

Description

Summary

OpenCode automatically injects the reasoningSummary parameter (likely defaulting to "auto" or similar) when calling GPT-5 series models (e.g., gpt-5, gpt-5-codex). This works for official OpenAI endpoints but breaks with custom OpenAI-compatible providers routed through LiteLLM (or Azure), where the backend models do not support this parameter.

This is a compatibility issue for users routing GPT-5-named models to third-party backends (e.g., DeepSeek, Gemini, or Azure deployments) via LiteLLM proxy.

Steps to Reproduce

Configure a custom OpenAI-compatible provider in opencode.jsonc:

"provider": {
  "compatible": {
    "npm": "@ai-sdk/openai-compatible",
    "options": {
      "baseURL": "https://your-litellm-proxy/v1",
      "apiKey": "{env:YOUR_KEY}"
    },
    "models": {
      "gpt-5-codex": { "name": "GPT-5 Codex" },
      // other GPT-5 series models
    }
  }
},
"model": "compatible/gpt-5-codex"

Plugins

None

OpenCode version

V1.1.65

Steps to reproduce

litellm.BadRequestError: litellm.ContentPolicyViolationError: litellm.ContentPolicyViolationError: AzureException - Unknown parameter: 'reasoningSummary'.
model=gpt-5. content_policy_fallback=None. fallbacks=None.

Screenshot and/or share link

No response

Operating System

Windows11

Terminal

Powershell

This issue might be related to existing issues about custom providers receiving unsupported parameters. Please check:

@ai-sdk/openai-compatible max_tokens error for GPT 5.x #5421: @ai-sdk/openai-compatible max_tokens error for GPT 5.x (similar parameter compatibility issue with custom providers)
Azure models fail due to reasoning_effort being sent even with --variant none #12113: Azure models fail due to reasoning_effort being sent even with --variant none (same pattern of reasoning parameters being sent to backends that don't support them)

These issues may help understand the broader parameter-handling pattern and could be relevant to the solution.

Could help to fix as of add a parameter to trun off/on the 'reasoningSummary'?

Having the same problem recently, using LiteLLM base URL with GPT-5.2...

Some logs:
litellm.BadRequestError: OpenAIException - Error code: 400 - {'path': '/ai-orchestration-api/v1/openai/chat/completions',
'history': [{'response': {'response': '{\n "error": {\n "message": "Unknown parameter: 'reasoning_summary'.",\n "type": "invalid_request_error",\n "param": "reasoning_summary",\n "code": "unknown_parameter"\n }\n}

Found this, maybe useful for a fix?
code-yeongyu/oh-my-openagent#1788
code-yeongyu/oh-my-openagent#1794

Same here, same logs (on Mac though):
litellm.BadRequestError: OpenAIException - Unknown parameter: 'reasoningSummary'.. Received Model Group=openai/gpt-5.2
Available Model Group Fallbacks=None

Had a similar issue with Claude:

litellm.BadRequestError: Vertex_aiException BadRequestError - b'{"type":"error","error":{"type":"invalid_request_error","message":"invalid beta flag"},"request_id":"..."}'. Received Model Group=claude-sonnet-4.5
Available Model Group Fallbacks=None

and now I have the same issue with GPT models

litellm.BadRequestError: litellm.ContentPolicyViolationError: litellm.ContentPolicyViolationError: AzureException - Unknown parameter: 'reasoningSummary'.
model=gpt-5.1. content_policy_fallback=None. fallbacks=None.
Set 'content_policy_fallback' - https://docs.litellm.ai/docs/routing#fallbacks. Received Model Group=gpt-5.1
Available Model Group Fallbacks=None

This worked for me (config in opencode.json) with v1.20.10:

"provider": {
    "my-litellm-provider": {
    "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://my-provider.url/api/v1"
      },
      "models": {
        "azure.gpt-5.1": {
          "name": "my-gpt-5.1",
          "options": {
            "reasoningSummary": null,
          }
        },
      }

      "models": {
        "azure.gpt-5.1": {
          "name": "my-gpt-5.1",
          "options": {
            "reasoningSummary": null,
          }
        },
      }

This does not work for me ~ on
opencode version 1.2.10
oh-my-opencode version 3.8.5

In my experience, setting it to null would likely not work, but setting it to false will, but at the expense of completely disabling reasoning.

Getting the same error when using Helicone.ai as my gateway for gpt models.

@ayaz

@ayaz Hmm, I just checked actually,
https://opencode.ai/docs/providers/#custom-provider
There is no other way to set a custom provider apart from using openai-compatible.

Some providers serve anthropic models, which actually support this parameter. The only way (workaround?) is to use openai-compatible

What are your thoughts in making this configurable? Any other suggestions?

#14783 (comment)

      "models": {
        "azure.gpt-5.1": {
          "name": "my-gpt-5.1",
          "options": {
            "reasoningSummary": null, // null or false is not work for me 
          }
        },
      }

opencode version 1.2.15

@aizigao

@aizigao That is because the moment you use strings like "gpt-5" in the model's ID or name, opencode considers it a GPT-5 model and starts sending the reasoning fields. The idea is to use a name for the id/name fields that does not include the GPT characters. Here's an example of what I used:

      "models": {
        "do-primary": {
          "id": "do-primary",
          "name": "GPT 5.2",
          "options": {
            "model": "openai-gpt-5.2"
          }
        }

@aizigao That is because the moment you use strings like "gpt-5" in the model's ID or name, opencode considers it a GPT-5 model and starts sending the reasoning fields. The idea is to use a name for the id/name fields that does not include the GPT characters. Here's an example of what I used:这是因为一旦你在模型的 ID 或名称中使用类似“gpt-5”这样的字符串，OpenCode 就会将其视为 GPT-5 模型并开始发送推理字段。因此，我们需要使用不包含 GPT 字符的 ID/名称字段。以下是我使用的示例：
      "models": {
        "do-primary": {
          "id": "do-primary",
          "name": "GPT 5.2",
          "options": {
            "model": "openai-gpt-5.2"
          }
        }

I resolve it by add a plugin

@aizigao That is because the moment you use strings like "gpt-5" in the model's ID or name, opencode considers it a GPT-5 model and starts sending the reasoning fields. The idea is to use a name for the id/name fields that does not include the GPT characters. Here's an example of what I used:
      "models": {
        "do-primary": {
          "id": "do-primary",
          "name": "GPT 5.2",
          "options": {
            "model": "openai-gpt-5.2"
          }
        }

I add a plugin to resolve it
add
~/.config/opencode/plugin/myArgPatch.ts

import { Plugin } from "@opencode-ai/plugin"

export const myArgPatch: Plugin = async (ctx) => {
  return {
    auth: {
      provider: "YUOR_PROVDER_NAME",
      loader: async (getAuth, provider) => {
        return {
          async fetch(input, init) {
            const opts = init ?? {}
            if (opts.body && typeof opts.body === "string") {
              try {
                const body = JSON.parse(opts.body)
     
                //  delete this arg 
                delete body.reasoningSummary
                opts.body = JSON.stringify(body)
              } catch (e) {}
            }
            return fetch(input, {
              ...opts,
              timeout: false,
            })
          },
        }
      },
    },
  }
}

in ~/config/opencode/opencode.jsonc

{
...
  "plugin": ["myArgPatch"],
}

+1. Still reproducible on the latest version (tested via Azure and custom provider).

I'm able to call GPT-5.4 with the setup below:

"gpt-5.4": {
  "name": "gpt-5.4",
  "limit": {
    "context": 1000000,
    "output": 128000
  },
  "options": {
    "reasoningSummary": null
  },
  "extra_body": {
    "drop_params": true
  }
},

I'm using opencode with version 1.2.27

I'm able to call GPT-5.4 with the setup below:

"gpt-5.4": {
"name": "gpt-5.4",
"limit": {
"context": 1000000,
"output": 128000
},
"options": {
"reasoningSummary": null
},
"extra_body": {
"drop_params": true
}
},
I'm using opencode with version 1.2.27

no joy -- still get reasoningSummary unknown parameter.

tried with previous example as follows and get an issue with max_tokens param:

"gpt-5.4": {
          "name": "gapt-5.4",
          "limit": {
            "context": 262144,
            "output": 25000
          },
          "modalities": {
            "input": ["text"],
            "output": ["text"]
          },
          "options": {
            "model":"gpt-5.4",
          }
        },

If you're using LiteLLM, adding a separate provider for OpenAI models (GPT-5.4) pointing at @ai-sdk/openai rather than @ai-sdk/openai-compatible should resolve the issue:

{
  "provider": {
    "my-litellm-provider": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://my-provider.url/api/v1"
      },
      "models": {
        ...
      }
    },
    "my-litellm-provider-openai": {
      "npm": "@ai-sdk/openai",
      "options": {
        "baseURL": "https://my-provider.url/api/v1"
      },
      "models": {
        ...
      }
    }
  }
}

If you are using the LiteLLM proxy, you can refer to this setting:

{
    "$schema": "https://opencode.ai/config.json",
    "provider": {
        "custom_provider": {
            "npm": "@ai-sdk/openai-compatible",
            "name": "custom_name",
            "options": { "baseURL": "${your_llm_base_url}" },
            "models": {
                "gpt-5-codex": { 
                    "name": "GPT-5 Codex",
                    "options": {
                        "reasoningSummary": null
                    }
                }
            }
        }
    }
}

"models": {
"azure.gpt-5.1": {
"name": "my-gpt-5.1",
"options": {
"reasoningSummary": null,
}
},
}
This does not work for me ~ on opencode version 1.2.10 oh-my-opencode version 3.8.5

Worked for me!! I'm using gpt-5.1 with variants and was facing the same problem, after adding the "options" part it fixed all of them :) Thanks!

If you're using LiteLLM, adding a separate provider for OpenAI models (GPT-5.4) pointing at @ai-sdk/openai rather than @ai-sdk/openai-compatible should resolve the issue:

{
"provider": {
"my-litellm-provider": {
"npm": "@ai-sdk/openai-compatible",
"options": {
"baseURL": "https://my-provider.url/api/v1"
},
"models": {
...
}
},
"my-litellm-provider-openai": {
"npm": "@ai-sdk/openai",
"options": {
"baseURL": "https://my-provider.url/api/v1"
},
"models": {
...
}
}
}
}

Worked for me ;)

I still have the error and I've tried all these options. I am not using LiteLLM. I think this might be related to this #14783

any fix regarding this issue guys? i still face same issue

@damarowen

@damarowen join discussion here:
#27716 (comment)

I managed to get this working with gpt5-4 in azure foundary using this:

"foundary2": { "name": "foundry-gpt", "npm": "@ai-sdk/openai", "options": { "baseURL": "https://foundry-123451.cognitiveservices.azure.com/openai/v1/" }, "models": { "gpt-5-4": { "name": "gpt-5-4", "family": "gpt-5", "reasoning": true, "interleaved": { "field": "reasoning_content" }, "modalities": { "input": ["text"], "output": ["text"] }, "temperature": true, "tool_call": true, "attachment": false, "experimental": false, "status": "active" } } }

Navigation Menu

Repository navigation

[Bug] Custom OpenAI-compatible provider errors with "Unknown parameter: 'reasoningSummary'" for GPT-5 series models #13546

Description

Summary

Steps to Reproduce

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

1 remaining item

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[Bug] Custom OpenAI-compatible provider errors with "Unknown parameter: 'reasoningSummary'" for GPT-5 series models #13546

Description

Description

Summary

Steps to Reproduce

Plugins

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

github-actions commented on Feb 14, 2026

Yang-Kevin-DH commented on Feb 14, 2026

acanova commented on Feb 15, 2026

zchrysaylor commented on Feb 19, 2026

X4ndras commented on Feb 20, 2026

ggbetz commented on Feb 24, 2026

shahrin014 commented on Feb 24, 2026

ayaz commented on Feb 25, 2026

dailyherold commented on Feb 25, 2026

shahrin014 commented on Feb 26, 2026

1 remaining item

aizigao commented on Mar 3, 2026

ayaz commented on Mar 3, 2026

aizigao commented on Mar 3, 2026

ar27111994 commented on Mar 14, 2026

billkuojhgf commented on Mar 16, 2026

absolutegravitas commented on Mar 19, 2026

paulgrow-octane commented on Mar 20, 2026

Yang-Kevin-DH commented on Mar 20, 2026

guilherme-deschamps commented on Mar 20, 2026

zeyd31 commented on Mar 24, 2026

royfrancis commented on Mar 24, 2026

damarowen commented on Jun 4, 2026

rekram1-node commented on Jun 4, 2026

hopoffbaby commented on Jun 8, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions