Skip to content

Add Infomaniak provider with 5 AI models#2

Open
pietrini wants to merge 5 commits into
devfrom
feature/add-infomaniak-provider
Open

Add Infomaniak provider with 5 AI models#2
pietrini wants to merge 5 commits into
devfrom
feature/add-infomaniak-provider

Conversation

@pietrini
Copy link
Copy Markdown
Owner

@pietrini pietrini commented May 6, 2026

Summary

This PR adds support for the Infomaniak AI provider, enabling access to multiple AI models through the Infomaniak API.

Key Changes

  • New provider: Added Infomaniak provider configuration with OpenAI-compatible API endpoint
  • New models: Added 5 AI models available through Infomaniak:
    • Ministral 3 14B Instruct 2512 (Mistral AI)
    • Apertus 70B Instruct 2509 (Swiss AI)
    • Kimi K2.6 (Moonshot AI)
    • Qwen3.5 122B-A10B FP8 (Alibaba)
    • Gemma 4 31B (Google)

Implementation Details

  • The Infomaniak provider uses the OpenAI-compatible SDK (@ai-sdk/openai-compatible)
  • Models are configured with their respective pricing (input/output tokens), context limits, and supported modalities
  • Some models extend existing model definitions from other providers while overriding Infomaniak-specific pricing
  • The provider requires two environment variables: INFOMANIAK_API_KEY and INFOMANIAK_PROJECT_ID
  • All models support text input, with Ministral 3 14B also supporting image input

claude added 5 commits May 6, 2026 18:17
Infomaniak is a Swiss cloud provider offering an OpenAI-compatible AI API
with European data sovereignty. Adds provider config and models:

- mistralai/Ministral-3-14B-Instruct-2512 (vision, tool calling)
- google/Gemma-4-31B-it (extends canonical, vision, reasoning)
- moonshotai/Kimi-K2.6 (extends canonical, reasoning, vision, video)
- Qwen/Qwen3.5-122B-A10B-FP8 (extends canonical, reasoning, multimodal)
- swiss-ai/Apertus-70B-Instruct-2509 (open multilingual model, beta)

https://claude.ai/code/session_01MQforMKhaKerRcDUMZjnVg
Infomaniak caps this model at 100,000 input tokens, lower than the
canonical Google model's 256K context window.

https://claude.ai/code/session_01MQforMKhaKerRcDUMZjnVg
Infomaniak only exposes text-to-text for this model, not the vision
capabilities available in the canonical Google model.

https://claude.ai/code/session_01MQforMKhaKerRcDUMZjnVg
Based on Infomaniak's model page screenshots:
- Qwen3.5 122B FP8: input cap 200K, image-text modality, beta
- Kimi K2.6: input cap 256K, image-text modality (no video), beta
- Ministral 3 14B: input cap 100K, beta
- Gemma 4 31B: beta
All non-Apertus models now correctly reflect Infomaniak's published specs.

https://claude.ai/code/session_01MQforMKhaKerRcDUMZjnVg
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.

2 participants