Models · Mistral AI · Out sinceReleased 2 Dec 2025
Ministral 3 14B
Ministral 3 14B is made by Mistral AI. We don't have enough test results yet to rank it. It's cheap to use. Its makers have published it, so you can run it on your own servers.
Dense 13.5B LM + 0.4B vision encoder, Apache 2.0, 256K context; FP8 instruct fits in 24GB VRAM (card). Base (Ministral-3-14B-Base-2512) and Reasoning variants published; also 8B and 3B sizes. OpenRouter ~$0.20/$0.20; Mistral list price not re-verified.
- —
- —
- —
- Cheap$0.20 / $0.20
- $0.20
- 262K
- —
- 9 (9 independent9 indep.)
- 2 Dec 2025
- Mistral AI
- text, image
Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.
You can't choose how long this one thinks.
No adjustable reasoning setting listed.
OpenRouter is a service that gives access to many AI models in one place. This model is listed there as mistralai/ministral-14b-2512.
Route it as mistralai/ministral-14b-2512 at $0.20 in / $0.20 out per 1M tokens, 262K context. Listed since 2 Dec 2025.
frequency_penaltymax_tokenspresence_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_p
Run it yourself
Self-hosting
On your own servers,Licence, sizeunder your own control.and hardware.
- Yes
- 14B
Runs on a powerful laptop or workstation.
≤ 24 GB at 4-bit (one consumer GPU / 32 GB Mac). Weights ≈ 15 GB at 8-bit, 8 GB at 4-bit (+10–30% for KV cache).
Languages: Languages: Dozens incl. en, fr, es, de, it, pt, nl, zh, ja, ko, ar (Swedish not listed). EuroEval Swedish: Base 1.99, Instruct 2.13.
Training it further: Fine-tuning: Card lists 'Fine-tuning and specialization' as a use case; base checkpoint published (no recipe in card).
Quantisations: fp8 (default release), bf16 (...-BF16), gguf (mistralai/Ministral-3-14B-Instruct-2512-GGUF). Engines: vLLM (recommended), Transformers.
Download (Hugging Face)Weights on Hugging Face ↗huggingface.co ↗huggingface.co ↗
Every test result
Every result
All the numbers,with where they came from.
Every result we've found for this model, with who measured it and a link to where we read it.
All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.
| NotesNotes | ||||||
|---|---|---|---|---|---|---|
| AA-LCR | 26.3% | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b |
| AA-Omniscience Index | -66.4 | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b; AA-Omniscience Index (-100..100) |
| Artificial Analysis Intelligence Index | 6 | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b |
| Artificial Analysis output speed | 73 tok/s | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b; median output tokens/sec across AA-tracked API providers (not a self-hosted measurement) |
| EuroEval Swedish (generative) | 2.1 | Defaultmistralai/Ministral-3-14B-Instruct-2512 (val) | Independent testIndependentEuroEval (Alexandra Institute) ↗ | 29 Sep 2026 | — | rank tier 10; ±0.19; lower is better; Base-2512 scores 1.99 ±0.14; SweDN 38.13, Skolprov 55.50, Swedish facts 21.00, ScaLA-sv 45.47 |
| GPQA Diamond | 57.2% | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b |
| Humanity's Last Exam | 4.6% | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b |
| SciCode | 23.8% | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b |
| Terminal-Bench 2.1 | 9.7% | DefaultMinistral 3 14B | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug ministral-3-14b |