feat: add Claude Opus 4.6 (fast mode) to GitHub Copilot provider

[mugnimaestra]

Summary

Adds the Claude Opus 4.6 (fast mode) (preview) model to the github-copilot provider.

This was announced on Feb 7, 2026 and is currently rolling out as a research preview for Copilot Pro+ and Enterprise users.

• Announcement: https://github.blog/changelog/2026-02-07-claude-opus-4-6-fast-is-now-in-public-preview-for-github-copilot/
• Supported models docs: https://docs.github.com/copilot/reference/ai-models/supported-models
• Related issue: Add Claude Opus 4.6 (fast mode) to GitHub Copilot provider #833

Notes & Caveats

I want to be transparent about a few uncertainties:

• Model ID: I've used claude-opus-4.6-fast as the filename/ID, following the pattern of grok-code-fast-1 already in this provider. However, I don't have access to the exact internal model ID that the Copilot API uses — this model is only visible to Pro+/Enterprise users in the model picker. If the maintainers know the correct ID, feel free to rename the file.
• Capabilities: I've mirrored the values from the existing claude-opus-4.6.toml since GitHub describes the fast mode as delivering "the same intelligence as Opus 4.6" but with up to 2.5x faster output token speeds.
• Status: This is a public preview model. I noticed other preview models in this provider (like gemini-3-flash-preview.toml) don't use the status field, so I've followed that pattern. Happy to add status = "beta" if preferred.
• Premium multiplier: 9x during the promotional period (through Feb 16, 2026). Not sure if this is tracked in the TOML schema.

Please let me know if anything needs adjusting — happy to update! 🙏

[nulladdict]

Currently model size definitions are a bit broken. If #810 lands you'll probably also need to update the limit section to:

context = 200_000
input = 128_000
output = 64_000

[rekram1-node]

Model is not available to OC users yet

[nkhoit]

If this helps i have access to the model and the metadata info from the API is here:

gh copilot -- --log-level debug -p "hi" -s --model claude-opus-4.6-fast
grep -A 50 'Got model info' ~/.copilot/logs/process-*.log

{
  "billing": {
    "is_premium": true,
    "multiplier": 9,
    "restricted_to": ["pro_plus", "enterprise"]
  },
  "capabilities": {
    "family": "claude-opus-4.6-fast",
    "limits": {
      "max_context_window_tokens": 200000,
      "max_non_streaming_output_tokens": 16000,
      "max_output_tokens": 64000,
      "max_prompt_tokens": 168000,
      "vision": {
        "max_prompt_image_size": 3145728,
        "max_prompt_images": 1,
        "supported_media_types": ["image/jpeg", "image/png", "image/webp"]
      }
    },
    "object": "model_capabilities",
    "supports": {
      "adaptive_thinking": true,
      "max_thinking_budget": 32000,
      "min_thinking_budget": 1024,
      "parallel_tool_calls": true,
      "streaming": true,
      "structured_outputs": true,
      "tool_calls": true,
      "vision": true
    },
    "tokenizer": "o200k_base",
    "type": "chat"
  },
  "id": "claude-opus-4.6-fast",
  "is_chat_default": false,
  "is_chat_fallback": false,
  "model_picker_category": "powerful",
  "model_picker_enabled": true,
  "name": "Claude Opus 4.6 (fast mode)",
  "object": "model",
  "policy": {
    "state": "enabled"
  }
}

[nkhoit]

Currently model size definitions are a bit broken. If #810 lands you'll probably also need to update the limit section to:

context = 200_000
input = 128_000
output = 64_000

From the API it says

...
      "max_context_window_tokens": 200000,
      "max_non_streaming_output_tokens": 16000,
      "max_output_tokens": 64000,
      "max_prompt_tokens": 168000,
...

So i guess the input got updated to 168000 recently?

[nkhoit]

Should just be "Claude Opus 4.6 (fast mode)", drop the (preview).
From the GHCP API response: #834 (comment)

[nkhoit]

realizing that on the CLI though it says (Preview) so there's a bit of discrepancy between what's in the API response and what's being shown in the model selector. Unsure what's the right thing to put here

[mugnimaestra]

thanks @nkhoit! updated both — dropped (preview) and bumped context to 200_000. though as @rekram1-node mentioned it's not available for OC users yet, so might just need to wait for now?