tinyanalysis

Leaderboard / Mistral AI

Ministral 3 3B

Mistral AI · Ministral 3 · released 2025-12-02 · data checked 2026-11-01

Ministral 3 3B is a 3B-parameter dense model from Mistral AI (Europe), released 2025-12-02 under the Apache-2.0 license. It scores 49.4 on the Small LM Index, #11 of 21 tiny models. It needs about 3.3 GB of memory at 4-bit with an 8K-token context and costs $0.10 per million tokens through the cheapest API provider.

Small LM Index
49.4
#11 of 21 scored tiny models
Likely 44.5–54.1 · from 3 of 6 core benchmarks
Parameters
3B
Dense
Price per 1M tokens
$0.10
$0.10 in / $0.10 out · blended 3:1
Context window
256K
tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 3 reported by Mistral AI. How we label sources.

GPQA Diamond Lab-reported Knowledge53.4
Class best: Nanbeige4.2 3B 87.4
AIME Lab-reported Math · AIME 2025 (Reasoning variant)72.1
Class best: Spark-X2.6 4B 91.2
LiveCodeBench Lab-reported Coding54.8
Class best: LFM3 1.7B 73.5

Not reported: estimated

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

MMLU-Pro≈ 72
Humanity's Last Exam≈ 6
SWE-bench Verified≈ 30

Specifications

Lab
Mistral AI (FR · Europe)
Architecture
Dense transformer
Total parameters
3B
Active parameters
All (dense)
Size class
Tiny
Context window
256K tokens
Input modalities
text, image
Reasoning
Hybrid (thinking on/off)
License
Apache-2.0
Openness
Open weights
Memory to run
~3.3 GB at 4-bit, 8K context (more sizes)
Hugging Face
17.7K downloads (30 days) · 142 likes
API providers
1 on OpenRouter
API price
$0.10 in / $0.10 out per 1M (OpenRouter, 2026-11-03)
Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Ranked in

Sources

Something wrong here? Report it.

Run it locally

Uses mistralai/Ministral-3-3B-Reasoning-2512-GGUF (Q4_K_M). Swap the tag for another size: BF16, Q4_K_M, Q5_K_M, Q8_0.

Ollama
ollama run hf.co/mistralai/Ministral-3-3B-Reasoning-2512-GGUF:Q4_K_M
llama.cpp
llama-server -hf mistralai/Ministral-3-3B-Reasoning-2512-GGUF:Q4_K_M
Then open http://localhost:8080 for a chat window.
vLLM (GPU server)
vllm serve mistralai/Ministral-3-3B-Reasoning-2512
Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization4K context32K context256K context
16-bit F167.0 GB10 GB34 GB
8-bit Q8_04.2 GB7.2 GB31 GB
4-bit Q4_K_M2.8 GB5.9 GB30 GB

Check it against your GPU, Mac or phone →