Allow selection of inference provider when using OpenRouter

I'm not aware of existing configuration options to configure a specific provider when using OpenRouter.

I think implementing a provider selector (similar to the model selector) would be a good QoL improvement (especially now to avoid low throughput kimi k2 providers)

The provider selector could be opened with a shortcut when an OpenRouter model is currently selected.

I don't know how much implementation effort this would result in, but I think it is manageable and doable through existing OpenRouter API endpoints:

https://openrouter.ai/docs/features/provider-routing
https://openrouter.ai/docs/api-reference/list-available-providers

I could work on this

I was able to make it work, thanks to this issue:

In order, you can put whatever provider is available in the model page. You can use options from provider options as well.

{
    "$schema": "https://opencode.ai/config.json",
    "provider": {
        "openrouter": {
            "options": {
                "name": "OpenRouter"
            },
            "models": {
                "moonshotai/kimi-k2": {
                    "name": "KIMI K2",
                    "options": {
                        "OpenRouter": {
                            "provider": {
                                "order": [
                                    "together/fp8",
                                    "baseten/fp4",
                                    "groq"
                                ],
                                "allow_fallbacks": false
                            }
                        }
                    }
                }
            }
        }
    }
}

I was able to make it work, thanks to this issue:

In order, you can put whatever provider is available in the model page. You can use options from provider options as well.

{
"$schema": "https://opencode.ai/config.json",
"provider": {
"openrouter": {
"options": {
"name": "OpenRouter"
},
"models": {
"moonshotai/kimi-k2": {
"name": "KIMI K2",
"options": {
"OpenRouter": {
"provider": {
"order": [
"together/fp8",
"baseten/fp4",
"groq"
],
"allow_fallbacks": false
}
}
}
}
}
}
}
}

If i only leave "groq" as an option I get:
AI_APICallError: No endpoints found for moonshotat/kimi-k2

Which doesn't really make sense. In the OpenRouter activity logs I can see that the request arrived, but ended with reason "stop"

Not sure what I'm doing wrong, but I can see that OpenRouter delegates execution to various providers despite explicitly instructed to use cerebras:

"openrouter": {
      "options": {
        "name": "OpenRouter"
      },
      "models": {
        "qwen/qwen3-235b-a22b-2507": {
          "name": "Custom-Q3-235b-cerebras",
          "options": {
            "OpenRouter": {
              "provider": {
                "order": [
                  "cerebras"
                ],
                "allow_fallbacks": false
              }
            }
          }
        }
      }

I suspect that params weren't actually passed to OpenRouter.

can confirm that with this config (limiting kimi-k2 to only groq) it does not work, while changing only groq -> baseten makes it work 🤷🏼‍♂️ :

{
  "$schema": "https://opencode.ai/config.json",
  "theme": "system",
  "model": "openrouter/moonshotai/kimi-k2",
  "autoshare": false,
  "autoupdate": false,
  "provider": {
    "openrouter": {
      "models": {
        "moonshotai/kimi-k2": {
          "options": {
            "provider": {
              "order": ["groq"],
              "allow_fallbacks": false
            }
          }
        }
      }
    }
  }
}

Here's the (trimmed and beautified) log:

ERROR 2025-08-01T16:16:13 +2ms service=session session=ses_799949420ffe4tW09KIBYWH16D error={
  "name": "AI_APICallError",
  "url": "https://openrouter.ai/api/v1/chat/completions",
  "requestBodyValues": {
    "model": "moonshotai/kimi-k2",
    "max_tokens": 32000,
    "temperature": 0,
    "provider": {
      "order": [
        "groq"
      ],
      "allow_fallbacks": false
    },
    "messages": [
      {
        "role": "system",
        "content": "You are opencode, an interactive CLI tool that ..."
      },
      {
        "role": "system",
        "content": "Here is some useful information about the environment you are running in: ..."
      },
      {
        "role": "user",
        "content": "hello"
      }
    ],
    "tools": [...],
    "tool_choice": "auto",
    "stream": true
  },
  "statusCode": 404,
  "responseHeaders": {
    "access-control-allow-origin": "*",
    "cf-ray": "9686a3812c9dc902-MAD",
    "connection": "keep-alive",
    "content-encoding": "gzip",
    "content-type": "application/json",
    "date": "Fri, 01 Aug 2025 16:16:13 GMT",
    "permissions-policy": "payment=(self \"https://checkout.stripe.com\" \"https://connect-js.stripe.com\" \"https://js.stripe.com\" \"https://*.js.stripe.com\" \"https://hooks.stripe.com\")",
    "referrer-policy": "no-referrer, strict-origin-when-cross-origin",
    "server": "cloudflare",
    "transfer-encoding": "chunked",
    "vary": "Accept-Encoding",
    "x-content-type-options": "nosniff"
  },
  "responseBody": "{\"error\":{\"message\":\"No endpoints found for moonshotai/kimi-k2.\",\"code\":404},\"user_id\":\"user_2t4PFDTP6kRNNXK1o4sIcg4b8Ib\"}",
  "isRetryable": false,
  "data": {
    "error": {
      "message": "No endpoints found for moonshotai/kimi-k2.",
      "code": 404
    }
  }
}

I've gotten it working for kimi k2 with groq:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "openrouter": {
        "moonshotai/kimi-k2": {
          "options": {
            "provider": {
              "only": ["groq"]
            },
            "max_tokens": 16384
          }
        }
      }
    }
  }
}

I found using only yielded better error messages (docs: https://openrouter.ai/docs/features/provider-routing#allowing-only-specific-providers), and it said "No allowed providers are available for the selected model". I then noticed in the error log given above that the max tokens were specified in the request, I checked models.dev and kimi k2 on openrouter has an output limit of 32,768 but groq only has 16384, so on a whim I changed it and it now works.

Additionally, I used the only param to try and force cerebras with Qwen3 Coder, but I got an error message about how it doesn't support tool calls (which tracks because the OR listing for cerebras Qwen3 Coder does not list tools as a support parameter).

In case this helps anyone - I found

using :nitro shortcut in front of the model name doesn't work
The max_tokens has to be <= to the max output tokens for that provider/model else it would refuse to run.

The below config worked for me, which I verified by looking at openrouter activity to see it actually used these providers.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "openrouter": {
      "models": {
        "qwen/qwen3-coder": {
          "name": "fast qwen3-coder",
          "options": {
            "provider": {
              "only": ["cerebras"]
            },
            "max_tokens": 32000
          }
        },
        "moonshotai/kimi-k2": {
          "name": "fast kimi k2",
          "options": {
            "provider": {
              "only": ["groq"]
            },
            "max_tokens": 16000
          }
        }
      }
    }
  }
}

Note: cerebras is not a reliable provider on openrouter - I found the speed greatly varies. groq is more consistently fast.

Looking at https://models.dev/?search=kimi-k2 - it lists providers, but when using a meta provider like openrouter there is no way to define the actual provider there.

@Arbyy

Thanks to @Arbyy at least it showed me an error message, instead of just nothing.

If someone still faces the same issue of empty responses even with Arbby config, try to use the full model's id like this "moonshotai/kimi-k2-0905" instead.

Is this issue related to why I don't see openrouter/auto in the list of models?

To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

It would be much better if the openrouter provider/plugin listed the provider prefixes, imho, as well as a * prefix, like */model-name for when you'll accept that model from any provider.

fix(web): stabilize Den onboarding flow with app routes (anomalyco#1010)

For future reference: was having the same issue as the others in Spain. Doing the same payment process through my phone instead (iPhone, Safari) had no issues whatsoever with the card details (no Apple Pay, though I assume that would have worked too).

Navigation Menu

Repository navigation

Allow selection of inference provider when using OpenRouter #1010

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Allow selection of inference provider when using OpenRouter #1010

Description

Activity

OneAbodFTW commented on Jul 15, 2025

nathabonfim59 commented on Jul 16, 2025

Flexinos commented on Jul 20, 2025

ivanbokii commented on Jul 30, 2025

grota commented on Aug 1, 2025

Arbyy commented on Aug 2, 2025

khalido commented on Aug 14, 2025

OneAbodFTW commented on Oct 30, 2025

niebloomj commented on Feb 15, 2026

github-actions commented on Apr 17, 2026

lee-b commented on May 10, 2026

cbossaers commented on Sep 1, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions