Mistral Small 3 vs Devstral Small 2 24B
Side-by-side specs, benchmarks, and a verdict by use case.
Updated 2026-07-13
| Spec | Mistral Small 3 | Devstral Small 2 24B |
|---|---|---|
| Parameters | 24B | 24B |
| Author | Mistral AI | Mistral AI |
| License | Apache 2.0 | Apache 2.0 |
| Context window | 0k | 0k |
| VRAM at Q4 | 14 GB | 14 GB |
| VRAM at Q5 | 17 GB | 17 GB |
| VRAM at Q8 | 26 GB | 26 GB |
| VRAM at FP16 | 48 GB | 48 GB |
| Use cases | chat, general, code | code, fr |
Verdict
Both models sit in a similar size class. The pick depends on tags, license, and benchmarks rather than raw parameter count.
The two models at a glance
About Mistral Small 3
Mistral AI's 24B dense model that closes most of the gap with 70B-class models. Best quality-per-parameter we've measured at this size in 2025. Strengths: Quality approaching Llama 3 70B at a third the size, Low latency relative to peers, 128k context window, Strong tool use and agent behavior.
About Devstral Small 2 24B
Mistral AI's 24B coding specialist co-developed with All Hands AI, scoring 72.2% on SWE-Bench under Apache 2.0. Fits on a single RTX 4090. Strengths: 72.2% SWE-Bench in a 24B dense model, Runs comfortably on a single RTX 4090, 256K context for whole-repo work, Apache 2.0 license.
How they compare
Mistral Small 3 comes from Mistral AI and Devstral Small 2 24B from Mistral AI. This comparison is built entirely from structured specs — parameter count, VRAM by quantization, context window, license, and published benchmark scores — so the verdict below reflects measurable differences rather than marketing claims.
Mistral Small 3 and Devstral Small 2 24B share the same 24B parameter class. Both need about 14 GB of VRAM at a Q4 quantization, so they fit the same GPU tier.
The two models target different sweet spots: Mistral Small 3 is tuned for chat, general, code, while Devstral Small 2 24B leans toward code, fr. Match the model to your dominant workload rather than to raw size.
For long-context work, Devstral Small 2 24B offers the bigger window (250k vs 32k tokens).
Memory, quantization & throughput
Across quantization levels, Mistral Small 3 requires Q4 ≈ 14 GB, Q5 ≈ 17 GB, Q8 ≈ 26 GB, FP16 ≈ 48 GB, while Devstral Small 2 24B requires Q4 ≈ 14 GB, Q5 ≈ 17 GB, Q8 ≈ 26 GB, FP16 ≈ 48 GB. In practice Mistral Small 3 needs a 16 GB card at Q4, so plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity of Q8 or FP16.
Without a GPU, Mistral Small 3 needs roughly 24 GB of system RAM to run on CPU and Devstral Small 2 24B about 24 GB — workable for offline use but far slower than GPU inference. On a mid-range GPU you can expect on the order of 15 tokens/sec from Mistral Small 3 and 15 from Devstral Small 2 24B, scaling up to 40 and 40 tokens/sec on high-end hardware.
Which fits your GPU
Here is the highest-quality quantization of each model that fits common GPU memory budgets, so you can match Mistral Small 3 or Devstral Small 2 24B to the card you actually own:
- On a 16 GB GPU: Mistral Small 3 runs at Q4 (14 GB); Devstral Small 2 24B runs at Q4 (14 GB).
- On a 24 GB GPU: Mistral Small 3 runs at Q5 (17 GB); Devstral Small 2 24B runs at Q5 (17 GB).
Benchmark scores
Reported benchmarks for Mistral Small 3: MMLU 81, GPQA 42.2, HumanEval 84.8.
Reported benchmarks for Devstral Small 2 24B: SWE-Bench 72.2.
Bottom line: which should you pick?
- Pick Devstral Small 2 24B for long-context work (up to 250k tokens).
- Pick Mistral Small 3 if your workload is chat, general.
- Pick Devstral Small 2 24B if your workload is fr.
Which GPU should you buy to run Mistral Small 3?
To run Mistral Small 3 locally at Q4, you need ~14 GB of VRAM. The best value for this is a RTX 5070 Ti (16 GB VRAM).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Frequently asked questions
What is the difference between Mistral Small 3 and Devstral Small 2 24B?
The headline differences: both are 24B models; their context windows differ (32k vs 250k tokens). Below we break down VRAM by quantization, benchmark scores, and a use-case verdict so you can pick the right one.
Can Mistral Small 3 and Devstral Small 2 24B run on a 24 GB GPU?
At a Q4 quantization, Mistral Small 3 needs about 14 GB of VRAM and fits comfortably on a 24 GB GPU; Devstral Small 2 24B needs about 14 GB and fits comfortably on a 24 GB GPU. Both have the same Q4 footprint.
What licenses do Mistral Small 3 and Devstral Small 2 24B use?
Mistral Small 3 is licensed under Apache 2.0 and Devstral Small 2 24B under Apache 2.0.
Which has the longer context window, Mistral Small 3 or Devstral Small 2 24B?
Devstral Small 2 24B has the larger context window (250k vs 32k tokens), so it handles longer documents and codebases in a single prompt.