Kolibri-1: Specs, License, Hardware & How to Run
Aleph Alpha German/English MoE reasoning model with tool calling and 1M context.
TL;DR. Kolibri-1 is a llm model from Aleph Alpha. Parameters: 78.1B (active 3.46B).License: Apache-2.0. It is listed under Closed LLM APIs.
| Developer | Aleph Alpha |
|---|---|
| Type | llm |
| Architecture | MoE |
| Parameters | 78.1B · active 3.46B |
| Modalities | text |
| Languages | German, English |
| Context | 1M |
| License | Apache-2.0 — Apache-2.0 as listed on the card; not legal advice. |
| Hardware | Not yet verified — VRAM and quantization options were not in the seed extract. (not independently tested) |
| Hugging Face likes | 908 (as of 2026-10-13) |
| Model card | https://huggingface.co/Aleph-Alpha/Kolibri-1 |
| First seen | 2026-10-02 |
Kolibri-1 is Aleph Alpha’s open MoE reasoning model with tool calling. We hang it off open alternatives to closed LLM APIs as an editorial mapping, not as a quality claim against any named closed vendor.
Specs (2026-10-13)
| Field | Value |
|---|---|
| Developer | Aleph Alpha |
| Total / active params | 78.1B / 3.46B |
| Architecture | MoE |
| Context | 1M |
| Languages | German, English |
| License | Apache-2.0 |
| Likes | 908 |
| Card date | 2026-10-02 |
We have no sourced benchmark table for this page. Hardware is unverified here.
Follow the official card for run instructions. forkalt has not published a tested serving guide for Kolibri-1 yet.
Deploy options
- local: Hugging Face model card
FAQ
How big is Kolibri-1?
78.1B total parameters, 3.46B active per token (MoE), according to the Hugging Face card checked 2026-10-13.
What is the context length?
1M, per the same card.
Which languages?
German and English are the languages stated in the verification note.
Can I use it commercially?
The card lists Apache-2.0. Read the license; this is not legal advice.
Sources & verification
License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.