Mistralai Devstral Small 2 24B Instruct 2512 GGUF Size & VRAM Requirements
mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF is a large instruction-tuned chat model focused on coding & software engineering. With approximately 24.01 billion parameters, it balances local hardware requirements with reasoning and generation fidelity. It is built on the mistral3 architecture with a native context window of up to 393,216 tokens. The model is primarily recommended for automated code completion, bug triage, unit test generation, refactoring, and multi-file codebases. Released under the apache-2.0 license, it can be executed fully offline without sending data to external APIs.
Under the hood it uses 40 transformer layers, a hidden size of 5,120, 32 attention heads. It uses grouped-query attention (32 query heads sharing 8 key/value heads), which already trims KV-cache memory compared with full multi-head attention.
To run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF locally at a 4,096-token context, its quantized versions need between 7.52 GB (IQ2_XXS, lowest quality) and 45.34 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF is about 0.62 GB, rising to roughly 60.0 GB at its full 393,216-token context. Shorter prompts free up memory for a higher-quality quantization.
For most users the best balance is IQ2_XXS, needing about 7.52 GB. That means bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF fits entirely in the VRAM of an 8 GB GPU or larger, running fully on the GPU.
Available GGUF quantizations for bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF include IQ2_XXS, IQ2_XS, IQ2_S, IQ2_M, Q2_K, IQ3_XXS, Q2_K_L, IQ3_XS, Q3_K_S, IQ3_M, Q3_K_M, Q3_K_L, IQ4_XS, Q3_K_XL, IQ4_NL, Q4_0, Q4_K_S, Q4_K_M, Q4_K_L, Q4_1, Q5_K_S, Q5_K_M, Q5_K_L, Q6_K, Q6_K_L, Q8_0, BF16. The model supports a native context length of up to 393,216 tokens; a longer context grows the KV cache and the memory needed.
Recommended Use Cases & Local Setup
Ideal for: Automated code completion, bug triage, unit test generation, refactoring, and multi-file codebases.
Direct integration into VS Code and JetBrains IDEs for inline tab completion
Command-line AI pair programmer that automatically applies Git diffs
Local OpenAI-compatible API endpoint serving your editor
GGUF file size and memory by quantization
Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.
| Quant. | Bits | Quality | Weights | KV | Total | Speed~ | Verdict |
|---|---|---|---|---|---|---|---|
| IQ2_XXS | 2.18 | Very low | 6.1 GB | 0.62 GB | 7.52 GB | 65.6 t/s | Fits in VRAM |
| IQ2_XS | 2.4 | Very low | 6.71 GB | 0.62 GB | 8.14 GB | 7.4 t/s | Offload |
| IQ2_S | 2.49 | Very low | 6.97 GB | 0.62 GB | 8.39 GB | 7.2 t/s | Offload |
| IQ2_M | 2.7 | Low | 7.56 GB | 0.62 GB | 8.98 GB | 6.6 t/s | Offload |
| Q2_K | 2.96 | Low | 8.28 GB | 0.62 GB | 9.71 GB | 6.0 t/s | Offload |
| IQ3_XXS | 3.09 | Low | 8.64 GB | 0.62 GB | 10.07 GB | 5.8 t/s | Offload |
| Q2_K_L | 3.18 | Low | 8.89 GB | 0.62 GB | 10.32 GB | 5.6 t/s | Offload |
| IQ3_XS | 3.3 | Fair | 9.23 GB | 0.62 GB | 10.65 GB | 5.4 t/s | Offload |
| Q3_K_S | 3.47 | Fair | 9.69 GB | 0.62 GB | 11.11 GB | 5.2 t/s | Offload |
| IQ3_M | 3.55 | Fair | 9.92 GB | 0.62 GB | 11.34 GB | 5.0 t/s | Offload |
| Q3_K_M | 3.82 | Fair | 10.69 GB | 0.62 GB | 12.11 GB | 4.7 t/s | Offload |
| Q3_K_L | 4.13 | Fair | 11.55 GB | 0.62 GB | 12.97 GB | 4.3 t/s | Offload |
| IQ4_XS | 4.25 | Good | 11.88 GB | 0.62 GB | 13.31 GB | 4.2 t/s | Offload |
| Q3_K_XL | 4.33 | Good | 12.1 GB | 0.62 GB | 13.52 GB | 4.1 t/s | Offload |
| IQ4_NL | 4.49 | Good | 12.54 GB | 0.62 GB | 13.97 GB | 4.0 t/s | Offload |
| Q4_0 | 4.5 | Good | 12.57 GB | 0.62 GB | 13.99 GB | 4.0 t/s | Offload |
| Q4_K_S | 4.51 | Good | 12.62 GB | 0.62 GB | 14.04 GB | 4.0 t/s | Offload |
| Q4_K_M | 4.78 | Good | 13.35 GB | 0.62 GB | 14.77 GB | 3.7 t/s | Offload |
| Q4_K_L | 4.94 | Good | 13.81 GB | 0.62 GB | 15.24 GB | 3.6 t/s | Offload |
| Q4_1 | 4.96 | Good | 13.85 GB | 0.62 GB | 15.28 GB | 3.6 t/s | Offload |
| Q5_K_S | 5.43 | Very good | 15.19 GB | 0.62 GB | 16.61 GB | 3.3 t/s | Offload |
| Q5_K_M | 5.59 | Very good | 15.61 GB | 0.62 GB | 17.04 GB | 3.2 t/s | Offload |
| Q5_K_L | 5.72 | Very good | 16.0 GB | 0.62 GB | 17.42 GB | 3.1 t/s | Offload |
| Q6_K | 6.45 | Very good | 18.02 GB | 0.62 GB | 19.44 GB | 2.8 t/s | Offload |
| Q6_K_L | 6.55 | Excellent | 18.32 GB | 0.62 GB | 19.75 GB | 2.7 t/s | Offload |
| Q8_0 | 8.35 | Excellent | 23.33 GB | 0.62 GB | 24.76 GB | — | Insufficient |
| BF16 | 15.71 | Excellent | 43.92 GB | 0.62 GB | 45.34 GB | — | Insufficient |
KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.
Frequently asked questions
What kind of model is bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF?
bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF is an instruction-tuned chat model with 24.01 billion parameters, based on the mistral3 architecture. It is released under the apache-2.0 license and distributed as GGUF files for local inference.
How much VRAM do you need to run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF?
You need about 7.52 GB of VRAM to run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF entirely on the GPU using the IQ2_XXS quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.
Can I run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF on an 8 GB GPU?
Yes. With 8 GB of VRAM you can run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF fully on the GPU using IQ2_XXS (about 7.52 GB).
Can I run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF on a 16 GB GPU?
Yes. With 16 GB of VRAM you can run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF fully on the GPU using Q4_1 (about 15.28 GB).
Can I run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF on a 24 GB GPU?
Yes. With 24 GB of VRAM you can run bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF fully on the GPU using Q6_K_L (about 19.75 GB).
What context length does bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF support?
bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF supports a native context length of up to 393,216 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.
What is the best quantization for bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF?
For bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF, a strong default is Q4_K_M, which needs about 14.77 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.