pinning OpenRouter provider doesn't work

Description

Configuration should pin the provider of LLM MiniMax-M 2.1 to Minimax:

			"minimax/minimax-m2.1": {
				"name": "MiniMax-M2.1",
				"options": {
					"provider": {
						"order": [
							"minimax"
						],
						"allow_fallbacks": false
					}
				}
			}

Unfortunately the usage log at openrouter.ai/activity shows the arbitrary providers of the MiniMax-M2.1 LLM are used.

Replacing "order" by "only" doesn't change the behavior.

Pinning to one specific provider seams not to work.

OpenCode version

1.1.34

Operating System

Arch Linux

Terminal

kitty

This issue might be a duplicate of existing issues. Please check:

Hidden calls Haiku #10272: Hidden calls Haiku (nearly identical - configured MiniMax M2.1 but Claude Haiku 4.5 appears in usage logs)
Openrouter unwated requests to Claude Haiku 4.5. #4579: Openrouter unwated requests to Claude Haiku 4.5 (same root cause with OpenRouter provider)
Custom OpenAI-compatible provider options not being passed to API calls #5674: Custom OpenAI-compatible provider options not being passed to API calls (provider configuration not respected)
Allow selection of inference provider when using OpenRouter #1010: Allow selection of inference provider when using OpenRouter (related feature request)
subagent provider/id parser is dead ? provider/model not found on subagent #8896: subagent provider/id parser is dead (provider configuration parsing issues)

Feel free to ignore if none of these address your specific case.

@thdxr

@thdxr Openrouter requests are going to random providers, this is kinda serious

I'm 100% sure OC doesn't include provider in the request

I chose amazon-bedrock in config and it shows Ncompass here:

"openai/gpt-oss-120b": {
    "name": "GPT OSS 120B",
    "options": {
        "OpenRouter": {
            "provider": {
                "order": [
                    "amazon-bedrock"
                ],
                "allow_fallbacks": false,
            }
        }
    }
}

Can we also have tests for this so it doesn't happen again, because sending your data to a random provider is pretty bad for privacy :(

If I set:

"openrouter": {
      "models": {
        "openai/gpt-oss-120b": {
          "name": "GPT OSS 120B",
          "options": {
            "provider": {
              "order": ["amazon-bedrock"],
              "allow_fallbacks": false,
            },
          },
        },
      },
    },

and look in the request body sent to openrouter, I see:

{
    "model": "openai/gpt-oss-120b",
    "max_tokens": 32000,
    "provider": {
        "order": [
            "amazon-bedrock"
        ],
        "allow_fallbacks": false
    },

.....
}

Is this incorrect?

the options under a model should be sent directly as is in the request body, unless it is a special ai sdk field

@rekram1-node

@rekram1-node

I see, i was wrong, looks like in my case, this was the issue:

    "options": {
        "OpenRouter": {
            "provider": {

there shouldn't have been a

  "OpenRouter": {

instead

    "options": {
        "provider": {

Thanks, this fixed it

But OP's config looks correct, and he still has issues

I tested op config, I can't replicate the issue, now that doesnt mean it doesnt exist but I don't see anything in our code that WOULDNT be sending the params as is

I tested op config, I can't replicate the issue, now that doesnt mean it doesnt exist but I don't see anything in our code that WOULDNT be sending the params as is

what is the correct config? "order" or "only"?
If I look at OpenRouters openrouter.ai/activity page for my account I see that varying providers are used for calling MiniMax-M2.1

@thdxr Openrouter requests are going to random providers, this is kinda serious

Came to report this.

Requests in the same session are going to all random providers - I just had a 100K tokens go out to 5 different providers... without caching. 😐💸

Needing to manually configure providers isn't a good default experience - one would hope the client is smart enough to choose wisely on it's own at the beginning of a session, e.g. the currently fastest available provider that supports caching, and then stick with the same provider for the session, in order to actually benefit from caching.

Please, this would be an amazing feature. Allow the user in the frontend of opencode to select a provider for openrouter.

I'm not sure how many people use OpenRouter, but it's my main, and having a random provider or having to go mess around with JSON for every new model kills the experience.

Hell, I can even code the feature.

@Syazvinski

@Syazvinski do you have any idea how to do it?

I pored over the OpenRouter docs, and I don't see any direct way to get the provider ID - response.metadata just seems to come back empty. Ironically, error responses have provider_name in the metadata, but successful responses don't seem to have this property - I can find nothing in the docs, or in the actual response body, or headers.

My best/only clue is the id returned with a response, but docs don't say what type of ID this is - there's zero documentation for this. There is another API here, which does return a provider_id, but for the same type input ID? You'd have to just try it and find out, but that's my best guess at the moment. Why is their documentation so bad. 😅

To others looking, here's how to pin your OpenRouter provider via opencode.jsonc:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "openrouter": {
      "models": {
        "minimax/minimax-m2.5": {
          "name": "Minimax M2.5 (official)",
          "options": {
            "provider": {
              "only": [
                "minimax/fp8"
              ]
            }
          }
        }
      }
    }
  }
}

I overpaid again today because I forgot I have to configure OpenCode on every system. 🤦‍♂️

Weird enough, but the property only in providr field fails json schema check

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

Navigation Menu

Repository navigation

pinning OpenRouter provider doesn't work #10557

Description

OpenCode version

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

pinning OpenRouter provider doesn't work #10557

Description

Description

OpenCode version

Operating System

Terminal

Activity

github-actions commented on Jan 25, 2026

TheHamkerCat commented on Jan 31, 2026

TheHamkerCat commented on Jan 31, 2026

rekram1-node commented on Jan 31, 2026

rekram1-node commented on Jan 31, 2026

TheHamkerCat commented on Jan 31, 2026

rekram1-node commented on Jan 31, 2026

olk commented on Feb 2, 2026

mindplay-dk commented on Feb 11, 2026

Syazvinski commented on Feb 14, 2026

mindplay-dk commented on Feb 18, 2026

mindplay-dk commented on Mar 8, 2026

XantreDev commented on May 12, 2026

github-actions commented on Jul 12, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions