Leaderboard / EuroLLM (UTTER consortium)

EuroLLM 22B Instruct

Name: EuroLLM 22B Instruct
Author: EuroLLM (UTTER consortium)

EuroLLM (UTTER consortium) · EuroLLM · released 2025-12-05 · data checked 2026-11-02

EuroLLM 22B Instruct is a 22.6B-parameter dense model from EuroLLM (UTTER consortium) (Europe), released 2025-12-05 under the Apache-2.0 license. It scores 19.2 on the Small LM Index, #34 of 34 medium models. It needs about 16.5 GB of memory at 4-bit with an 8K-token context.

Covers all 24 official EU languages. Scores read from the results tables (images) in the model card; the lab ran every model on its own harness, without long reasoning.

Small LM Index

19.2

#34 of 34 scored medium models

Likely 15.5–23.7 · from 2 of 6 core benchmarks

Parameters

22.6B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

32K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 6 reported by EuroLLM (UTTER consortium). How we label sources.

MMLU-Pro Lab-reported Knowledge · Instruct model, lab’s own harness (all scores)50.9

Class best: Qwen3.6 27B 86.2

GPQA Diamond Lab-reported Knowledge · GPQA Diamond26.8

Class best: Nanbeige4.3 33B 89.5

MMLU Lab-reported Knowledge69.8

Class best: EXAONE 4.0 32B 92.3

Multilingual MMLU Lab-reported Multilingual · Translated MMLU, averaged over EuroLLM languages64.1

Class best: EXAONE 4.0 32B 85.6

IFEval Lab-reported Instruction following67.2

Class best: LFM2.5 24B-A2B 96.8

MATH-500 Lab-reported Math54.5

Class best: Ternary Bonsai 2 27B 98.8

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 1

AIME≈ 15

LiveCodeBench≈ 19

SWE-bench Verified≈ 5

Specifications

Lab: EuroLLM (UTTER consortium) (PT · Europe)
Architecture: Dense transformer
Total parameters: 22.6B
Active parameters: All (dense)
Size class: Medium
Context window: 32K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: Apache-2.0
Openness: Fully open: weights, training data and code
Memory to run: ~16.5 GB at 4-bit, 8K context (more sizes)
Hugging Face: 8.5K downloads (30 days) · 88 likes

Weights on Hugging Face ↗

Compare with

EuroLLM 22B Instruct vs gpt-oss-20b (20.9B, index 64.0)
EuroLLM 22B Instruct vs LFM2.5 24B-A2B (24B, index 61.4)
EuroLLM 22B Instruct vs Magistral Small 1.2 (24B, index 55.9)
EuroLLM 22B Instruct vs Tri-21B Think (20.7B, index 43.6)
EuroLLM 22B Instruct vs Mistral Small 3.2 (24B, index 38.6)

Ranked in

Sources

Something wrong here? Report it.

Run it locally

Uses bartowski/utter-project_EuroLLM-22B-Instruct-2512-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ2_M, IQ2_S, IQ2_XS, IQ3_M, IQ3_XS, IQ3_XXS, IQ4_NL….

Ollama

ollama run hf.co/bartowski/utter-project_EuroLLM-22B-Instruct-2512-GGUF:Q4_K_M

llama.cpp

llama-server -hf bartowski/utter-project_EuroLLM-22B-Instruct-2512-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/EuroLLM-22B-Instruct-2512-mlx-4bit

vLLM (GPU server)

vllm serve utter-project/EuroLLM-22B-Instruct-2512

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context
16-bit F16	47 GB	53 GB
8-bit Q8_0	26 GB	32 GB
4-bit Q4_K_M	16 GB	22 GB

Check it against your GPU, Mac or phone →