Update gemma-4-31b to gemma-4-31b-it that’s the damn correct name

[immapolar]

Description

Every time I send a prompt in opencode with Gemma 4 selected, I get this crap:

models/gemma-4-31b is not found for API version v1beta, or is not supported for generateContent. Call ListModels to see available models.

That’s because the damn name should be models/gemma-4-31b-it, not models/gemma-4-31b (try a curl on ListModels, you’ll see). gemma-4-31b-it works in Goose, but Goose is ass; opencode is better.

Also, opencode ignores any config paths or files I try to force it to use gemma-4-31b-it; it sticks with god damn gemma-4-31b.

Plugins

None

OpenCode version

1.3.15

Steps to reproduce

Common sense.

Screenshot and/or share link

Operating System

6.6.87.2-microsoft-standard-WSL2

Terminal

Windows Terminal

[pchuck]

When you say opencode ignores your config paths or files, can you post an excerpt from your opencode.json where you tried to specify models/gemma-4-31b-it instead of models/gemma-4-31b, and of the curl listmodels output (presumably on the google v1beta api endpoint)?

[nik-55]

Facing Same error "models/gemma-4-31b is not found for API version v1beta, or is not supported for generateContent. Call ListModels to see available models"

Steps to reproduce

• Run /connect , select Google and then select Gemma 4 26B
• When we chat, it throw above error

[immapolar]

When you say opencode ignores your config paths or files, can you post an excerpt from your opencode.json where you tried to specify models/gemma-4-31b-it instead of models/gemma-4-31b, and of the curl listmodels output (presumably on the google v1beta api endpoint)?

I nuked all the config files so unfortunately I can’t really attach a JSON.

As for the Google v1beta API endpoint, here it is:

 curl "https://generativelanguage.googleapis.com/v1beta/models?key=..." \
    -H 'Content-Type: application/json' \
    -X GET
{
  "models": [
    {
      "name": "models/gemma-4-26b-a4b-it",
      "version": "001",
      "displayName": "Gemma 4 26B A4B IT",
      "description": "Gemma 4 26B A4B IT",
      "inputTokenLimit": 262144,
      "outputTokenLimit": 32768,
      "supportedGenerationMethods": ["generateContent","countTokens"],
      "temperature": 1,
      "topP": 0.95,
      "topK": 64,
      "maxTemperature": 2,
      "thinking": true
    },
    {
      "name": "models/gemma-4-31b-it",
      "version": "001",
      "displayName": "Gemma 4 31B IT",
      "description": "Gemma 4 31B IT",
      "inputTokenLimit": 262144,
      "outputTokenLimit": 32768,
      "supportedGenerationMethods": ["generateContent","countTokens"],
      "temperature": 1,
      "topP": 0.95,
      "topK": 64,
      "maxTemperature": 2,
      "thinking": true
    },
    ...
  ]
}

[pchuck]

what happens when you force it with explicit config

~/.config/opencode/opencode.json

{
    "$schema": "https://opencode.ai/config.json",
    "provider": {
        "google": {
            "name": "google",
            "options": {
                "baseURL": "https://generativelanguage.googleapis.com/v1beta",
            },
            "models": {
                "models/gemma-4-26b-a4b-it": {
                    "name": "models/gemma-4-26b-a4b-it (google)",
                }
            }
        }
    }
}

~/.local/share/opencode/auth.json

{
  "google": {
    "type": "api",
    "key": "..."
  }
}

then pick that model in the models drop-down, rather than the one autodetected from the provider models api call. (your path to these config files will be different on Windows)

[gc1o1]

Editing config.json by hand worked.
It's just annoying that we have to do that. I hope opencode next version fixes it.
Thanks!

[pchuck]

cool. just a casual user/observer/wannabe contributor, here. glad it worked. yeah, sounds like the bug should stay open.

[immapolar]

cool. just a casual user/observer/wannabe contributor, here. glad it worked. yeah, sounds like the bug should stay open.

Tried your opencode.json (for Windows folks: %USERPROFILE%/.config/opencode/opencode.json, create it if it’s missing) and yeah… it actually worked. Cheers 👍

Hoping the next version fixes the bug

[subhamgiri460]

what happens when you force it with explicit config

~/.config/opencode/opencode.json

{
"$schema": "https://opencode.ai/config.json",
"provider": {
"google": {
"name": "google",
"options": {
"baseURL": "https://generativelanguage.googleapis.com/v1beta",
},
"models": {
"models/gemma-4-26b-a4b-it": {
"name": "models/gemma-4-26b-a4b-it (google)",
}
}
}
}
}
~/.local/share/opencode/auth.json

{
"google": {
"type": "api",
"key": "..."
}
}
then pick that model in the models drop-down, rather than the one autodetected from the provider models api call. (your path to these config files will be different on Windows)

it worked thanks

[BrutchsamaJeanLouis]

there is also issue 16531 that blocks gemma4 support