Ministral-3-8B-Instruct-2512
Where it runs 8K context · ≥ 5 t/s floor
| Setup | Best plan | Memory | Generate |
|---|---|---|---|
| 8 GB GPU + 32 GB RAM | Q4_K_Mllama.cpp CUDAFull GPU | Fits | ~28–41 t/s Estimated |
| 12 GB GPU + 32 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~29–31 t/s Estimated |
| 16 GB GPU + 64 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~59–73 t/s Estimated |
| 24 GB GPU + 64 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~60–75 t/s Estimated |
| 32 GB GPU + 128 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~94–151 t/s Estimated |
| 2× 24 GB GPUs + 128 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~55–71 t/s Estimated |
| 96 GB workstation GPU + 256 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~97–152 t/s Estimated |
| Mac, 24 GB | Q8_0llama.cpp MetalUnified memory | Fits | ~20–24 t/s Estimated |
| Mac, 48 GB | Q8_0llama.cpp MetalUnified memory | Fits | ~20–24 t/s Estimated |
| Mac, 128 GB | Q8_0llama.cpp MetalUnified memory | Fits | ~33–50 t/s Estimated |
| Mac, 512 GB | Q8_0llama.cpp MetalUnified memory | Fits | ~37–56 t/s Estimated |
| DGX Spark 128 GB | Q8_0llama.cpp CUDAUnified memory | Fits | ~22–24 t/s Estimated |
| Ryzen AI Max+ 395, 128 GB | Q8_0llama.cpp VulkanUnified memory | Fits | ~15–21 t/s Estimated |
| CPU only, 64 GB DDR5-5600 dual channel | Q4_K_Mllama.cpp CPUCPU only | Fits | ~3.5–4.2 t/s Estimated extrapolated |
Same plan engine as the planner, with its default reserves and RAM. How plans are chosen.
Files exact Hugging Face bytes
| Artifact | Size | Files | Revision | Source |
|---|---|---|---|---|
| Q8_0 mistralai/Ministral-3-8B-Instruct-2512-GGUF | 8.41 GiB | 1 | 0102285ad7 | HF files |
| Q5_K_M mistralai/Ministral-3-8B-Instruct-2512-GGUF | 5.64 GiB | 1 | 0102285ad7 | HF files |
| Q4_K_M mistralai/Ministral-3-8B-Instruct-2512-GGUF | 4.84 GiB | 1 | 0102285ad7 | HF files |
| Native safetensors mixed/custom native; quant_method=fp8; stored tensor dtypes= mistralai/Ministral-3-8B-Instruct-2512 | 9.70 GiB | 3 | 5b26027e7b | HF files |
Not found in the representative quant search (not proven unavailable): GGUF FP16, MLX 8-bit, MLX 4-bit, EXL3, AWQ, GPTQ. Complete selected file sets; projectors, MTP/draft and lookup files are listed separately and budgeted only by plans that use them. File bytes are not allocated memory.
Quality
| AA evaluated mode | Index | LiveCodeBench | Terminal-Bench 2.1 | GPQA | HLE | Source |
|---|---|---|---|---|---|---|
| Ministral 3 8B | 5.5 | 25.7% | 4.1% | 47.1% | 4.3% | source |
Arena text rating
Unknown
Publisher-reported benchmarks
None extracted from the model card.
AA and Arena describe the hosted model in the stated mode, not each quantized file; scores are never inherited by base, pruned, abliterated or quantized variants and never averaged together.
How to run it
No model-specific recipe in the research. The planner binds the backend’s documented command template (llama.cpp, Ollama, LM Studio, MLX, vLLM) to your plan: open a plan’s details.
Recipes are source examples (settings are the publisher’s or author’s, not measured optima) and were not executed by LocalFit. A downloadable file does not prove your backend build supports the architecture.
Reported speeds 0
No reported measurement yet. Submit one.