Leaderboard / Aleph Alpha
Kolibri-1
Aleph Alpha · Kolibri · released 2026-10-03 · data checked 2026-11-01
Kolibri-1 is a 78.1B-parameter Mixture-of-Experts model (3.5B active per token) from Aleph Alpha (Europe), released 2026-10-03 under the Apache-2.0 license. It scores 68.3 on the Small LM Index, #2 of 6 large models. It needs about 49.6 GB of memory at 4-bit with an 8K-token context.
Bilingual English/German MoE (78B total, 3.5B active) with a 1M-token context.
Small LM Index
68.3
#2 of 6 scored large models
Likely 66.8–69.8 · from 6 of 6 core benchmarks
Parameters
78.1B
3.5B active · MoE
Price per 1M tokens
—
No hosted API on OpenRouter; self-host
Context window
1M
tokens
Benchmarks
Each bar shows this model's score. The tick marks the best large model on that benchmark. Sources: 10 reported by Aleph Alpha. How we label sources.
AIME Lab-reported Math · AIME 202696.0
LiveCodeBench Lab-reported Coding · LiveCodeBench v685.9
Terminal-Bench Lab-reported Agentic · Terminal-Bench 2.127.7
Class best: Nemotron 3.5 Bolt 74B-A9B 53.9
τ²-Bench Lab-reported Agentic · Average of Telecom, Retail and Airline80.4
IFBench Lab-reported Instruction following · Loose prompt accuracy78.1
BFCL Lab-reported Agentic · BFCL v461.4
Specifications
- Lab
- Aleph Alpha (DE · Europe)
- Architecture
- Mixture-of-Experts
- Total parameters
- 78.1B
- Active parameters
- 3.5B
- Size class
- Large
- Context window
- 1M tokens
- Input modalities
- text
- Reasoning
- Reasoning (thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~49.6 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 22.8K downloads (30 days) · 1K likes
Compare with
- Kolibri-1 vs Nemotron 3.5 Bolt 74B-A9B (74B, index 73.3)
- Kolibri-1 vs Qwen3-Next 80B-A3B Thinking (80B, index 63.3)
- Kolibri-1 vs DeepSeek-R1-Distill-Llama-70B (70.6B, index 50.1)
- Kolibri-1 vs Llama 3.3 70B (70.6B, index 40.8)
- Kolibri-1 vs K2-V2 Instruct (70B, index 57.9)
Ranked in
- #7 in Best small open models for coding
- #1 in Best small open models for math
- #1 in Best European open language models
Sources
Something wrong here? Report it.
Run it locally
No GGUF build from a well-known publisher yet.
vLLM (GPU server)
vllm serve Aleph-Alpha/Kolibri-1Full-precision weights, for a server GPU.
Memory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 256K context |
|---|---|---|---|
| 16-bit F16 | 158 GB | 159 GB | 164 GB |
| 8-bit Q8_0 | 85 GB | 86 GB | 90 GB |
| 4-bit Q4_K_M | 50 GB | 50 GB | 55 GB |