Skip to content

Add model settings for Claude 4.x family on Azure AI Foundry#5168

Open
Cyberfilo wants to merge 1 commit into
Aider-AI:mainfrom
Cyberfilo:feat/add-azure-ai-claude-models
Open

Add model settings for Claude 4.x family on Azure AI Foundry#5168
Cyberfilo wants to merge 1 commit into
Aider-AI:mainfrom
Cyberfilo:feat/add-azure-ai-claude-models

Conversation

@Cyberfilo
Copy link
Copy Markdown
Contributor

Summary

Adds four entries to `aider/resources/model-settings.yml` for the Anthropic Claude 4.x family on Azure AI Foundry (`azure_ai/` provider — distinct from the Azure OpenAI `azure/` provider):

Model weak_model_name editor_model_name
`azure_ai/claude-opus-4-7` `azure_ai/claude-haiku-4-5` `azure_ai/claude-sonnet-4-6`
`azure_ai/claude-opus-4-6` `azure_ai/claude-haiku-4-5` `azure_ai/claude-sonnet-4-6`
`azure_ai/claude-sonnet-4-6` `azure_ai/claude-haiku-4-5` `azure_ai/claude-sonnet-4-6`
`azure_ai/claude-haiku-4-5` `azure_ai/claude-haiku-4-5` `azure_ai/claude-haiku-4-5`

These models already have pricing/context-window metadata in litellm's `model_prices_and_context_window.json` (the upstream Aider pulls from at `ModelInfoManager.MODEL_INFO_URL`), but were missing the project-side settings. Without them, users routing Claude through Azure AI Foundry got the wrong edit grammar and no `cache_control` wiring.

Settings shape

Mirrors the existing native Anthropic / Vertex AI / OpenRouter Claude entries:

  • `edit_format: diff`
  • `cache_control: true`
  • `use_repo_map: true`
  • `max_tokens: 128000` for Opus, `64000` for Sonnet/Haiku
  • `accepts_settings: ["thinking_tokens"]` where the native variant has it (Opus 4.6, Sonnet 4.6, Haiku 4.5)
  • `use_temperature: false` + `overeager: true` for Opus 4.7, matching the native `claude-opus-4-7` block

Test plan

  • YAML validity: `python -c "import yaml; yaml.safe_load(open('aider/resources/model-settings.yml'))"` parses cleanly.
  • Diff is purely additive: 60 lines in model-settings.yml + 1 line in HISTORY.md.

Litellm's pricing JSON tracks the Anthropic Claude 4.x family on Azure
AI Foundry (azure_ai/claude-opus-4-7, azure_ai/claude-opus-4-6,
azure_ai/claude-sonnet-4-6, azure_ai/claude-haiku-4-5) but Aider was
missing the project-side settings entirely. Without explicit settings,
users hitting these models via Azure AI Foundry got the wrong edit
grammar and no cache-control wiring.

Adds four settings entries mirroring the corresponding native
Anthropic / OpenRouter / Vertex AI Claude entries:

- edit_format: diff
- weak_model_name routes to azure_ai/claude-haiku-4-5
- cache_control enabled
- max_tokens 128000 for Opus, 64000 for Sonnet/Haiku
- accepts_settings: ["thinking_tokens"] where the native variant has it
  (Opus 4.6, Sonnet 4.6, Haiku 4.5; Opus 4.7 uses use_temperature: false
  to match the native pattern instead)

Also adds a HISTORY.md entry under the main branch section.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant