Ministral 3 14B suits teams that want open-weight reasoning without renting a large proprietary model: the reasoning variant scores 85.0% on AIME 2025 per Mistral's model card, and Apache 2.0 licensing lets you self-host it freely. Skip it for real-time voice work, since there is no native audio input.
What changed
- · Release Ministral 3 14B released — 85% AIME 2025, strongest open-source 14B reasoning (ministral-3-14b-2512) source
Ministral 3 14B is Mistral AI's open-weight multimodal model announced on December 2, 2025; its reasoning variant scores 85% on AIME 2025 according to Mistral. It reads text and images through an embedded vision encoder, outputs text only, and costs $0.20 per million tokens.
Where it sits
- $0.2/M$ per 1M tokensBlended price (3:1)Lower is better#15 / 98peer median $1.70/Mvendor price, checked by Hokai
- 62 tok/stokens/sOutput speedHigher is better#38 / 55peer median 94 tok/scited: Artificial Analysis
- 71.2%% correctGPQA DiamondHigher is better#52 / 63peer median 87.8%per source, see benchmark scores
Cheaper than 82% of the 98 GA models with a published price, in the bottom third on GPQA Diamond (rank 52 of 63), and one of 87 whose vendor states it does not train on customer data.
Ranks are against GA models on Hokai that publish the same figure; ties share a rank.
Provider: Mistral AI · Family: Ministral 3
More about Mistral AI on Hokai
Context window: 256,000 tokens
Input modalities: text, image, tool-calls · Output: text, tool-calls
About Ministral 3 14B
Ministral 3 14B is the largest model in Mistral AI's Ministral 3 family, announced December 2, 2025 under the Apache 2.0 license. Paris-based Mistral AI ships it in three variants on Hugging Face: a base model, an instruct model, and a dedicated reasoning model, all distilled from the same recipe.
The dense Transformer architecture combines a 13.5 billion parameter language decoder with a 410 million parameter Vision Transformer encoder, for roughly 14 billion parameters total. Grouped Query Attention across 40 layers with a 5,120 hidden dimension keeps memory use in check without the added complexity of a Mixture-of-Experts design.
On AIME 2025, the reasoning variant scores 85.0%, against 73.7% for Qwen3-14B (Thinking) in the comparison table on Mistral's own Hugging Face model card. The same table lists 71.2% on GPQA Diamond for Ministral 3 14B and 66.3% for Qwen3-14B (Thinking). Both figures are Mistral's own, so treat them as vendor-reported until a third party reproduces them.
The 256,000 token context window matches Mistral Large 3. Inputs are text and image via the embedded vision encoder; outputs are text only, with function calling and structured JSON output support for agentic workflows. Audio and video are not supported.
In BF16 the model needs 32GB of VRAM; FP8 brings that down to 24GB for a single RTX 4090 or RTX 3090, and Q4 quantization drops it under 10GB for smaller GPUs. Weights are on Hugging Face and the model is also available through NVIDIA Build. Smaller siblings, Ministral 3 8B and Ministral 3 3B, suit tighter hardware.
Who should skip it: teams that need the best multi-step agent reasoning are better served by Mistral Large 3, and anyone who needs voice input must add a separate speech-to-text step. Artificial Analysis lists the non-reasoning version at 62.3 output tokens per second, slower than the 85 it quotes as average.
Alignment follows SFT plus RLHF on the instruct variant, with lighter default refusals than many proprietary chat models; Mistral expects deployers to add their own content policies for consumer-facing use. API inputs sent through la Plateforme are not used for training, and self-hosted deployments have no external data exposure.
Pricing
Ministral 3 14B costs $0.20 per 1 million input tokens and $0.20 per 1 million output tokens on Mistral's API. Cached input is billed at $0.02 per 1 million tokens, a tenth of the standard input rate. Self-hosted deployments pay only their own compute cost.
Prices checked on Mistral's docs pricing page on 2026-10-30. For scale, the smaller Ministral 3 8B is billed at $0.15 per 1M tokens each way, while Mistral Large 3 charges $0.50 input and $1.50 output.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.0060 | $0.0002 | $0.0062 |
| Support reply | $0.0004 | $0.0001 | $0.0005 |
| One coding agent run | $0.040 | $0.0040 | $0.044 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- 85% on AIME 2025 (Reasoning Variant): The dedicated reasoning model ID scores 85% on the AIME 2025 math competition benchmark, Mistral's highest published result in the Ministral 3 family.
- 71.2% on GPQA Diamond (Reasoning Variant): The same reasoning model ID scores 71.2% on GPQA Diamond, a graduate-level science question set, per Mistral's own Hugging Face model card.
- $0.02 Cached Input Rate: Cached input tokens cost $0.02 per 1M on Mistral's API, a tenth of the $0.20 standard input rate, per the docs pricing page.
- 256K Context Window: Supports a 256,000 token context window using grouped-query attention, matching Mistral Large 3's window at a fraction of the parameter count.
- 24GB FP8 Self-Hosting: Runs in FP8 on a single RTX 4090 or 3090 with 24GB of VRAM, the largest Ministral 3 variant that fits on one consumer flagship card.
- Apache 2.0 with Three Variants: Base, instruct, and reasoning variants are released under Apache 2.0, Mistral's most permissive license, enabling unrestricted commercial self-hosting.
Pros
- Strong reasoning benchmark results at the 14B parameter class, competitive with much larger proprietary models on math and science tasks.
- Runs on a single consumer flagship GPU once quantized to FP8, unlike larger open-weight reasoning models that need multi-GPU setups.
- Includes native vision input in the same model checkpoint, avoiding a separate OCR or captioning pipeline some competitors require.
Cons
- The full-precision BF16 weights need more VRAM than a single consumer flagship GPU provides; a quantized build is required to self-host on one card.
- The reasoning benchmark results apply to a separate model ID from the standard instruct variant, so teams must route math and logic tasks to the right endpoint.
- No native audio input, so voice applications need a separate speech-to-text step before calling the model.
Benchmarks
- AIME 2025: 85% vendor-reported · 30 Oct 2026 — Competition-level maths problems from the 2025 exam, % solved.
- GPQA Diamond: 71.2% vendor-reported · 30 Oct 2026 — PhD-level science questions that are hard to search for, % correct.
- AA blended price: $0.2/M cited: Artificial Analysis · 30 Oct 2026 — Price per 1M tokens at a 3:1 input to output blend, as listed by Artificial Analysis.
- Output speed: 62 tok/s cited: Artificial Analysis · 30 Oct 2026 — Median tokens written per second as measured by Artificial Analysis.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much does Ministral 3 14B cost in 2026?
Mistral bills Ministral 3 14B at $0.20 per million input tokens and $0.20 per million output tokens, with cached input at $0.02 per million. Running the open weights yourself sidesteps per-token billing, leaving just your own hardware or cloud compute bill.
Does Ministral 3 14B have a free plan?
Mistral's pricing page lists no free allowance for Ministral 3 14B on the paid API. The only no-cost path is self-hosting the Apache 2.0 weights on your own hardware, which avoids per-token charges but requires a GPU capable of running the model.
What are Ministral 3 14B's closest competitors?
Open-weight models in the same size class include Qwen3-14B and Google's Gemma 4 12B. Inside Mistral's own lineup, Ministral 3 8B is cheaper per token and Mistral Large 3 is the pick when you need stronger multi-step reasoning.
How does Ministral 3 14B compare with Qwen3-14B?
On Mistral's own Hugging Face model card, the Ministral 3 14B reasoning variant scores 85.0% on AIME 2025 and 71.2% on GPQA Diamond, against 73.7% and 66.3% for Qwen3-14B (Thinking). Those are vendor-reported numbers, and Ministral 3 14B adds a native vision encoder under the Apache 2.0 license.
How do you get started with Ministral 3 14B?
Create an API key in Mistral's Studio console and call the ministral-3-14b-reasoning-2512 model ID for math and logic tasks, or the instruct ID for chat and summarization. Self-hosting needs vLLM or a similar inference server and enough VRAM for FP8 weights, or a Q4 build for smaller GPUs.