Ministral 3 family (3B / 8B / 14B)

Mistral's clean Apache-2.0 edge family with Base / Instruct / Reasoning splits per size. The "no-license-drama" alternative to Qwen or Gemma when lawyers are involved.

License: Apache 2.0 (full family, all variants) · Context: 256K · Released: December 2, 2025

The decision in five lines

The call: Buy — for chat
Best for: chat · docs · agents
Runs on: 23 hardware picks fit (cheapest: Intel Arc B580 12 GB · $249)
Watch out: When you need the absolute top of Arena at 8–14B — Qwen 3.5 9B and Gemma 4 26B edge it on benchmarks.
Evidence: Estimated · last verified April 2026

3B: PARAMETERS
DENSE: TYPE
256K: CONTEXT
~2 GB (3B) / ~5 GB (8B) / ~9 GB (14B): VRAM AT Q4

Where we recommend this

Every tier slot in the planner where this model is a top or alternate pick. Pulled live from planner.js — when the planner refreshes, this table stays current.

CHAT · HIGH

Ministral 3 14B InstructMistral 14B with 256K + vision + tool use; Apache 2.0. Prefer Instruct — community reports timeouts on the Reasoning variant.

CHAT · MID

Ministral 3 8B InstructOutperforms Gemma 12B in most evals; Apache 2.0.

CHAT · LOW

Ministral 3 3BSmallest Ministral; reasoning + tool use; Apache 2.0.

DOCS · HIGH

Ministral 3 14B Instruct (256K context)256K context in a compact dense model; use Instruct — Reasoning variant has community-reported timeouts.

DOCS · MID

Ministral 3 8B Instruct + RAGSolid 8B + RAG; Mistral vocab handles docs cleanly.

AGENTS · HIGH

Ministral 3 14B InstructDense with strong tool use + planning. Prefer Instruct — community reports timeouts on Reasoning variant.

AGENTS · MID

Ministral 3 8B InstructTool use + reasoning; Apache 2.0.

AGENTS · LOW

Ministral 3 3BSmallest Ministral with reasoning + tool use.

The call

Mistral's clean Apache-2.0 edge family with Base / Instruct / Reasoning splits per size. The "no-license-drama" alternative to Qwen or Gemma when lawyers are involved.
When not to use: When you need the absolute top of Arena at 8–14B — Qwen 3.5 9B and Gemma 4 26B edge it on benchmarks. Also: Reasoning variant has community-reported timeouts on long chains; stick to Instruct for agents.

Runner notes

Ollama uses hyphen: `ministral-3:3b` / `ministral-3:8b` / `ministral-3:14b`. Instruct vs Reasoning is a separate tag — confirm which you pulled. 14B FP8 fits 24 GB.

License: Apache 2.0 (full family, all variants)
Released: December 2, 2025
Maker: Mistral AI
Model card: huggingface.co/mistralai/Ministral-3-14B-Instruct-2512 →

Hardware that fits

Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.

Next step

Find-by-model — see what hardware runs this→