Kolibri-1
A German-English open-weight MoE from Aleph Alpha, pitched on European AI sovereignty
Context window
262K (validated to 1M)
Input / 1M tokens
Free
Output / 1M tokens
Free
Provider
Aleph Alpha
Open weights under Apache 2.0 on Hugging Face (FP8 and BF16). No public API pricing; Aleph Alpha offers commercial deployment via sales. · Data verified 2026-10-08
Kolibri-1 (October 3, 2026) is an open-weight Mixture-of-Experts model from German lab Aleph Alpha with 78.1B total and 3.46B active parameters (384 experts, 6 per token). It is released under Apache 2.0 with a native 262K context validated up to 1M tokens. It launched the same week as Mistral Large 4, giving Europe two open-weight contenders.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Download the weights from Hugging Face (Aleph-Alpha/Kolibri-1). Aleph Alpha recommends serving at up to 262K tokens of context.
Strengths
- ✓Apache 2.0 license with weights available now
- ✓Very low active parameter count (3.46B) for cheap self-hosting
- ✓Strong German and English performance
- ✓Long context: 262K native, validated to 1M
Best for developers who...
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| AIME 2025 (English) | 96.9 | Aleph Alpha-reported; German 87.5 |
| GPQA Diamond | 84.3 | Aleph Alpha-reported |
| LiveCodeBench v6 | 85.9 | Aleph Alpha-reported |
Source: Hugging Face model card
Compare Kolibri-1
Compare Kolibri-1 with any other model
Build a comparison →All model comparisons →