Abliteration parameters
Table with columns: Parameter, Value| Parameter | Value |
|---|
| direction_index | per layer |
| attn.o_proj.max_weights.0 | 0: 0.85 |
| attn.o_proj.max_weights.1 | 1: 1.08 |
| attn.o_proj.max_weights.2 | 2: 1.16 |
| attn.o_proj.max_weights.3 | 3: 1.18 |
| attn.o_proj.max_weights.4 | 4: 1.28 |
| attn.o_proj.max_weights.5 | 5: 1.28 |
| attn.o_proj.max_weights.6 | 6: 1.42 |
| attn.o_proj.max_weights.7 | 7: 0.83 |
| attn.o_proj.max_weight_position | 31.80 |
| attn.o_proj.min_weights.0 | 0: 0.67 |
| attn.o_proj.min_weights.1 | 1: 0.85 |
| attn.o_proj.min_weights.2 | 2: 0.55 |
| attn.o_proj.min_weights.3 | 3: 0.98 |
| attn.o_proj.min_weights.4 | 4: 0.22 |
| attn.o_proj.min_weights.5 | 5: 0.25 |
| attn.o_proj.min_weights.6 | 6: 0.60 |
| attn.o_proj.min_weights.7 | 7: 0.71 |
| attn.o_proj.min_weight_distance | 9.10 |
| mlp.down_proj.max_weights.0 | 0: 1.36 |
| mlp.down_proj.max_weights.1 | 1: 0.85 |
| mlp.down_proj.max_weights.2 | 2: 0.89 |
| mlp.down_proj.max_weights.3 | 3: 1.31 |
| mlp.down_proj.max_weights.4 | 4: 1.44 |
| mlp.down_proj.max_weights.5 | 5: 1.18 |
| mlp.down_proj.max_weights.6 | 6: 0.98 |
| mlp.down_proj.max_weights.7 | 7: 1.44 |
| mlp.down_proj.max_weight_position | 27.71 |
| mlp.down_proj.min_weights.0 | 0: 1.05 |
| mlp.down_proj.min_weights.1 | 1: 0.36 |
| mlp.down_proj.min_weights.2 | 2: 0.49 |
| mlp.down_proj.min_weights.3 | 3: 0.77 |
| mlp.down_proj.min_weights.4 | 4: 1.34 |
| mlp.down_proj.min_weights.5 | 5: 0.87 |
| mlp.down_proj.min_weights.6 | 6: 0.14 |
| mlp.down_proj.min_weights.7 | 7: 0.04 |
| mlp.down_proj.min_weight_distance | 6.47 |
Table with columns: Metric, This model, Original model (mistralai/Ministral-3-14B-Instruct-2512-BF16)| Metric | This model | Original model (mistralai/Ministral-3-14B-Instruct-2512-BF16) |
|---|
| KL divergence | 0.0828 | 0 (by definition) |
| Refusals | 1/100 | 99/100 |
Ministral 3 14B Instruct 2512 BF16
The largest model in the Ministral 3 family, Ministral 3 14B offers frontier capabilities and performance comparable to its larger Mistral Small 3.2 24B counterpart. A powerful and efficient language model with vision capabilities.
This model is the instruct post-trained version, fine-tuned for instruction tasks, making it ideal for chat and instruction based use cases.
The Ministral 3 family is designed for edge deployment, capable of running on a wide range of hardware. Ministral 3 14B can even be deployed locally, capable of fitting in 32GB of VRAM in BF16, and less than 24GB of RAM/VRAM when quantized.
We provide a no-loss FP8 version here, you can find other formats and quantizations in the Ministral 3 - Additional Checkpoints collection.
Learn more in our blog post and paper.
Key Features
Ministral 3 14B consists of two main architectural components:
- 13.5B Language Model
- 0.4B Vision Encoder
The Ministral 3 14B Instruct model offers the following capabilities:
- Vision: Enables the model to analyze images and provide insights based on visual content, in addition to text.
- Multilingual: Supports dozens of languages, including English, French, Spanish, German, Italian, Portuguese, Dutch, Chinese, Japanese, Korean, Arabic.
- System Prompt: Maintains strong adherence and support for system prompts.
- Agentic: Offers best-in-class agentic capabilities with native function calling and JSON outputting.
- Edge-Optimized: Delivers best-in-class performance at a small scale, deployable anywhere.
- Apache 2.0 License: Open-source license allowing usage and modification for both commercial and non-commercial purposes.
- Large Context Window: Supports a 256k context window.
Use Cases
Private AI deployments where advanced capabilities meet practical hardware constraints:
- Private/custom chat and AI assistant deployments in constrained environments
- Advanced local agentic use cases
- Fine-tuning and specialization
- And more...
Bringing advanced AI capabilities to most environments.
Ministral 3 Family
Table with columns: Model Name, Type, Precision, Link| Model Name | Type | Precision | Link |
|---|
| Ministral 3 3B Base 2512 | Base pre-trained | BF16 | Hugging Face |
| Ministral 3 3B Instruct 2512 | Instruct post-trained | BF16 | Hugging Face |
| Ministral 3 3B Reasoning 2512 | Reasoning capable | BF16 | |
Other formats available here.
Benchmark Results
We compare Ministral 3 to similar sized models.
Reasoning
Table with columns: Model, AIME25, AIME24, GPQA Diamond, LiveCodeBench| Model | AIME25 | AIME24 | GPQA Diamond | LiveCodeBench |
|---|
| Ministral 3 14B | 0.850 | 0.898 | 0.712 | 0.646 |
| Qwen3-14B (Thinking) | 0.737 | 0.837 | 0.663 | 0.593 |
| | | | |
|
Instruct
Table with columns: Model, Arena Hard, WildBench, MATH Maj@1, MM MTBench| Model | Arena Hard | WildBench | MATH Maj@1 | MM MTBench |
|---|
| Ministral 3 14B | 0.551 | 68.5 | 0.904 | 8.49 |
| Qwen3 14B (Non-Thinking) | 0.427 | 65.1 | 0.870 | NOT MULTIMODAL |
| Gemma3-12B-Instruct | 0.436 | 63.2 | 0.854 | 6.70 |
Base
Table with columns: Model, Multilingual MMLU, MATH CoT 2-Shot, AGIEval 5-shot, MMLU Redux 5-shot, MMLU 5-shot, TriviaQA 5-shot| Model | Multilingual MMLU | MATH CoT 2-Shot | AGIEval 5-shot | MMLU Redux 5-shot | MMLU 5-shot | TriviaQA 5-shot |
|---|
| Ministral 3 14B | 0.742 | 0.676 | 0.648 | 0.820 | 0.794 | 0.749 |
| Qwen3 14B Base | 0.754 | 0.620 | 0.661 | 0.837 | 0.804 |
License
This model is licensed under the Apache 2.0 License.
You must not use this model in a manner that infringes, misappropriates, or otherwise violates any third party’s rights, including intellectual property rights.