Mistral 7B Instruct v0.3 Older
Where it runs 8K context · ≥ 5 t/s floor
| Setup | Best plan | Memory | Generate |
|---|---|---|---|
| 8 GB GPU + 32 GB RAM | Q5_K_Mllama.cpp CUDAFull GPU | Fits | ~29–42 t/s Estimated |
| 12 GB GPU + 32 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~33–36 t/s Estimated |
| 16 GB GPU + 64 GB RAM | Q8_0llama.cpp CUDAFull GPU | Fits | ~69–85 t/s Estimated |
| 24 GB GPU + 64 GB RAM | FP16llama.cpp CUDAFull GPU | Fits | ~38–48 t/s Estimated |
| 32 GB GPU + 128 GB RAM | FP16llama.cpp CUDAFull GPU | Fits | ~60–96 t/s Estimated |
| 2× 24 GB GPUs + 128 GB RAM | FP16llama.cpp CUDAFull GPU | Fits | ~35–45 t/s Estimated |
| 96 GB workstation GPU + 256 GB RAM | FP16llama.cpp CUDAFull GPU | Fits | ~62–97 t/s Estimated |
| Mac, 24 GB | FP16llama.cpp MetalUnified memory | Fits | ~13–15 t/s Estimated |
| Mac, 48 GB | FP16llama.cpp MetalUnified memory | Fits | ~13–15 t/s Estimated |
| Mac, 128 GB | FP16llama.cpp MetalUnified memory | Fits | ~21–32 t/s Estimated |
| Mac, 512 GB | FP16llama.cpp MetalUnified memory | Fits | ~23–36 t/s Estimated |
| DGX Spark 128 GB | FP16llama.cpp CUDAUnified memory | Fits | ~14–15 t/s Estimated |
| Ryzen AI Max+ 395, 128 GB | FP16llama.cpp VulkanUnified memory | Fits | ~9.3–14 t/s Estimated |
| CPU only, 64 GB DDR5-5600 dual channel | Q4_K_Mllama.cpp CPUCPU only | Fits | ~4.1–4.9 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 |
|---|---|---|---|---|
| FP16 MaziyarPanahi/Mistral-7B-Instruct-v0.3-GGUF | 13.50 GiB | 1 | ce89f59575 | HF files |
| Q8_0 MaziyarPanahi/Mistral-7B-Instruct-v0.3-GGUF | 7.17 GiB | 1 | ce89f59575 | HF files |
| Q5_K_M MaziyarPanahi/Mistral-7B-Instruct-v0.3-GGUF | 4.78 GiB | 1 | ce89f59575 | HF files |
| Q4_K_M MaziyarPanahi/Mistral-7B-Instruct-v0.3-GGUF | 4.07 GiB | 1 | ce89f59575 | HF files |
| AWQ 4-bit solidrust/Mistral-7B-Instruct-v0.3-AWQ | 3.88 GiB | 1 | 95b1295ddd | HF files |
| Native safetensors BF16 mistralai/Mistral-7B-Instruct-v0.3 | 13.50 GiB | 3 | c170c708c4 | HF files |
Not found in the representative quant search (not proven unavailable): MLX 8-bit, MLX 4-bit, EXL3, 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
No Artificial Analysis evaluation mapped to this checkpoint.
Arena text rating
Unknown
Publisher-reported benchmarks
None extracted from the model card.
Open LLM Leaderboard (historical): 19.23 average · dataset. Historical Open LLM Leaderboard row (dataset last modified 2025-03-20); not 2026 coverage.
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 61
| Hardware | Quant | Backend | Placement | Prompt t/s | Gen t/s | Date | Source |
|---|---|---|---|---|---|---|---|
| ASUS NVIDIA GeForce RTX 5060 Ti 16GB | Q8_0 | llama.cpp Vulkan | Full GPU | 3,681 | 53 | 2026-10-05 | openbenchmarking.org |
| ASUS NVIDIA GeForce RTX 5060 Ti 16GB | Q8_0 | llama.cpp CUDA | Full GPU | 3,958 | 53 | 2026-10-05 | openbenchmarking.org |
| Sapphire AMD Radeon AI PRO R9700 | Q8_0 | llama.cpp Vulkan | Full GPU | 4,438 | 75 | 2026-05-13 | openbenchmarking.org |
| AMD Radeon PRO W7900 | Q8_0 | llama.cpp Vulkan | Full GPU | 2,718 | 80 | 2026-05-11 | openbenchmarking.org |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96GB | Q8_0 | llama.cpp CUDA | Full GPU | 16,755 | 178 | 2026-05-03 | openbenchmarking.org |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96GB | Q8_0 | llama.cpp Vulkan | Full GPU | 14,262 | 171 | 2026-05-03 | openbenchmarking.org |
| NVIDIA RTX 6000 Ada Generation 48GB | Q8_0 | llama.cpp CUDA | Full GPU | 8,239 | 104 | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 6000 Ada Generation 48GB | Q8_0 | llama.cpp Vulkan | Full GPU | 6,648 | 103 | 2026-05-02 | openbenchmarking.org |
| Gigabyte NVIDIA GeForce RTX 5090 32GB | Q8_0 | llama.cpp CUDA | Full GPU | 14,836 | 177 | 2025-11-02 | openbenchmarking.org |
| Zotac NVIDIA GeForce RTX 5080 16GB | Q8_0 | llama.cpp CUDA | Full GPU | — | 105 | 2025-10-27 | openbenchmarking.org |
| AMD Radeon RX 9070 XT | Q8_0 | llama.cpp Vulkan | Full GPU | 2,988 | 72 | 2025-09-06 | openbenchmarking.org |
| ASUS NVIDIA GeForce RTX 5090 32GB | Q8_0 | llama.cpp CUDA | Full GPU | 12,860 | 166 | 2025-01-25 | openbenchmarking.org |
| ASUS NVIDIA GeForce RTX 5060 Ti 16GB | Q8_0 | llama.cpp Vulkan | Full GPU | 3,574 | — | 2026-10-05 | openbenchmarking.org |
| ASUS NVIDIA GeForce RTX 5060 Ti 16GB | Q8_0 | llama.cpp CUDA | Full GPU | 3,844 | — | 2026-10-05 | openbenchmarking.org |
| Dell Pro Max with GB10 (NVIDIA GB10, 128GB; DGX Spark OEM) | Q8_0 | llama.cpp CUDA | Full GPU | — | 30 | 2026-06-21 | openbenchmarking.org |
| Dell Pro Max with GB10 (NVIDIA GB10, 128GB; DGX Spark OEM) | Q8_0 | llama.cpp CUDA | Full GPU | 2,904 | — | 2026-06-21 | openbenchmarking.org |
| Sapphire AMD Radeon AI PRO R9700 | Q8_0 | llama.cpp Vulkan | Full GPU | 4,256 | — | 2026-05-13 | openbenchmarking.org |
| AMD Radeon PRO W7600 | Q8_0 | llama.cpp Vulkan | Full GPU | 817 | 34 | 2026-05-12 | openbenchmarking.org |
| AMD Radeon PRO W7600 | Q8_0 | llama.cpp Vulkan | Full GPU | 766 | — | 2026-05-12 | openbenchmarking.org |
| AMD Radeon PRO W7700 | Q8_0 | llama.cpp Vulkan | Full GPU | 1,523 | 54 | 2026-05-11 | openbenchmarking.org |
| AMD Radeon PRO W7700 | Q8_0 | llama.cpp Vulkan | Full GPU | 1,459 | — | 2026-05-11 | openbenchmarking.org |
| AMD Radeon PRO W7900 | Q8_0 | llama.cpp Vulkan | Full GPU | 2,567 | — | 2026-05-11 | openbenchmarking.org |
| Framework Desktop, AMD Ryzen AI Max+ 395, 64GB (4 x 16GB LPDDR5-8000), Radeon 8060S multi-device | Q8_0 | llama.cpp Vulkan | Layer split across GPUs | 1,170 | 28 | 2026-05-07 | openbenchmarking.org |
| Framework Desktop, AMD Ryzen AI Max+ 395, 64GB (4 x 16GB LPDDR5-8000), Radeon 8060S multi-device | Q8_0 | llama.cpp Vulkan | Layer split across GPUs | 1,090 | — | 2026-05-07 | openbenchmarking.org |
| Intel Arc Pro B50 BMG G21 | Q8_0 | llama.cpp Vulkan | Full GPU | 843 | 9.0 | 2026-05-06 | openbenchmarking.org |
| Intel Arc Pro B50 BMG G21 | Q8_0 | llama.cpp Vulkan | Full GPU | 674 | — | 2026-05-06 | openbenchmarking.org |
| Intel Arc B580 BMG G21 | Q8_0 | llama.cpp Vulkan | Full GPU | 1,696 | 15 | 2026-05-05 | openbenchmarking.org |
| Intel Arc B580 BMG G21 | Q8_0 | llama.cpp Vulkan | Full GPU | 1,257 | — | 2026-05-05 | openbenchmarking.org |
| Intel BMG G31 | Q8_0 | llama.cpp Vulkan | Full GPU | 2,029 | 21 | 2026-05-04 | openbenchmarking.org |
| Intel BMG G31 | Q8_0 | llama.cpp Vulkan | Full GPU | 1,581 | — | 2026-05-04 | openbenchmarking.org |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96GB | Q8_0 | llama.cpp CUDA | Full GPU | 16,229 | — | 2026-05-03 | openbenchmarking.org |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96GB | Q8_0 | llama.cpp Vulkan | Full GPU | 13,918 | — | 2026-05-03 | openbenchmarking.org |
| NVIDIA RTX 2000 Ada Generation 16GB | Q8_0 | llama.cpp CUDA | Full GPU | 2,142 | 27 | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 2000 Ada Generation 16GB | Q8_0 | llama.cpp CUDA | Full GPU | 2,081 | — | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 2000 Ada Generation 16GB | Q8_0 | llama.cpp Vulkan | Full GPU | 1,841 | 27 | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 2000 Ada Generation 16GB | Q8_0 | llama.cpp Vulkan | Full GPU | 1,792 | — | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 4000 Ada Generation 20GB | Q8_0 | llama.cpp CUDA | Full GPU | 3,904 | 42 | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 4000 Ada Generation 20GB | Q8_0 | llama.cpp CUDA | Full GPU | 3,795 | — | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 4000 Ada Generation 20GB | Q8_0 | llama.cpp Vulkan | Full GPU | 3,244 | 43 | 2026-05-02 | openbenchmarking.org |
| NVIDIA RTX 4000 Ada Generation 20GB | Q8_0 | llama.cpp Vulkan | Full GPU | 3,165 | — | 2026-05-02 | openbenchmarking.org |
Reported by the linked sources, not reproduced by LocalFit. Stream pieces/s and method-unknown rates are flagged and cannot pass a tokens/s floor. Showing 40 of 61. All rows