[FEATURE]: Grok 4.2 models

[v1gnesh]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

What's needed to add support for grok 4.2 models that are available to some via public beta?

[github-actions]

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

• [FEATURE]:x-ai/grok-4.1-fast model availability #4987: Similar feature request for adding a newer xAI Grok model to OpenCode

[v1gnesh]

Related - https://github.com/xai-org/xai-sdk-python/pull/105/files

[thoughtentity]

So I use the following config to get the Grok 4.20 models partially working. Unfortunately, the mult-agent model requires /responses OpenAi endpoint, which is hard-coded to specific providers in OpenCode.

X-ai needs a specific entry in the CUSTOM_LOADERS map in provider.ts to use the multi-agent model.

xai: async () => {
  return {
    autoload: false,
    async getModel(sdk: any, modelID: string) {
      return sdk.responses(modelID)
    },
    options: {},
  }
},

Provider Config

{
  "provider": {
    "xai": {
      "npm": "@ai-sdk/xai",
      "models": {
        "grok-4.20-multi-agent-experimental-beta-0304": {
          "name": "Grok 4.20 Multi-Agent Beta 0304",
          "family": "grok",
          "attachment": true,
          "reasoning": true,
          "tool_call": false,
          "temperature": true,
          "cost": {
            "input": 2,
            "output": 6,
            "cache_read": 0.2,
          },
          "limit": {
            "context": 2000000,
            "output": 30000,
          },
          "interleaved": {
            "field": "reasoning_content",
          },
          "variants": {
            "low-4-agents": {
              "reasoning": {
                "effort": "low"
              },
            },
            "medium-4-agents": {
              "reasoning": {
                "effort": "medium"
              },
            },
            "high-16-agents": {
              "reasoning": {
                "effort": "high"
              },
            },
            "xhigh-16-agents": {
              "reasoning": {
                "effort": "xhigh"
              },
            },
          },
        },
        "grok-4.20-experimental-beta-0304-reasoning": {
          "name": "Grok 4.20 Beta 0304",
          "family": "grok",
          "attachment": true,
          "reasoning": true,
          "tool_call": true,
          "temperature": true,
          "cost": {
            "input": 2,
            "output": 6,
            "cache_read": 0.2,
          },
          "limit": {
            "context": 2000000,
            "output": 30000,
          },
          "interleaved": {
            "field": "reasoning_content",
          },
          "variants": {
            "low": {
              "reasoning": {
                "effort": "low",
              },
            },
            "medium": {
              "reasoning": {
                "effort": "medium",
              },
            },
            "high": {
              "reasoning": {
                "effort": "high",
              },
            },
            "xhigh": {
              "reasoning": {
                "effort": "xhigh",
              },
            },
          },
        },
        "grok-4.20-experimental-beta-0304-non-reasoning": {
          "name": "Grok 4.20 Beta 0304 (Non-Reasoning)",
          "family": "grok",
          "attachment": true,
          "reasoning": false,
          "tool_call": true,
          "temperature": true,
          "cost": {
            "input": 2,
            "output": 6,
            "cache_read": 0.2,
          },
          "limit": {
            "context": 2000000,
            "output": 30000,
          },
        },
      },
    },
  },
}

[alexyaroshuk]

i think when it comes to adding models/providers, should refer to https://github.com/anomalyco/models.dev and make a PR there

[thoughtentity]

Per the docs sent to early access users, these grok model names are NOT stable. Additionally, OpenCode code changes are needed to fully enable the Grok 4.20 multi-agent models due to the OpenAi responses endpoint being hard-coded to specific provider.

[v1gnesh]

Thanks @alexyaroshuk, added here - anomalyco/models.dev#1114

[Bortus-AI]

Hopefully support soon. I love 4.2 multi-agent

[v1gnesh]

Above PR merged, so closing this.
Will have to work out multi-agent support next/separately.