Measured model
Measured on a Mac Studio (M5 Max, 36GB)Devstral-Small-2-24B-Instruct-2512
Mistral · Ollama name hf.co/unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF:Q4_K_M
On a Mac Studio (M5 Max, 36GB), Devstral-Small-2-24B-Instruct-2512 answers a simple question in about 0.2 s, gets 66% of the Japanese knowledge questions right, 82% of the math problems, and uses 16.3GB of memory.
These numbers are from before the method changed on Oct 2, 2026 (see how it was measured).
Results
| Measure | Result | Among measured models | Note |
|---|---|---|---|
| Reply time | about 0.2 s | 6 of 30 | Until one simple common-sense question is answered (median of 50) |
| Knowledge (high school to university) | 66% | 25 of 30 | 200 questions, ±7 |
| Math (word problems) | 82% | 14 of 30 | 100 questions, ±9, 1 cut off (counted wrong) |
| Memory used | 16.3GB | 17 of 30 | |
| EN→JA translation | 33 pts | 7 of 28 | out of 100 (chrF), about 1.6 s a sentence |
| JA→EN translation | 57 pts | 8 of 28 | out of 100 (chrF), about 1 s a sentence |
"Among measured models" is the place among the 30 models measured on this site (lower is better for time and memory).
Try it on your Mac
- Install the Mac version of Ollama from ollama.com and start it.
- Type this in Terminal. The first time, the model downloads; then you can chat with it.
ollama run hf.co/unsloth/Devstral-Small-2-24B-Instruct-2512-GGUF:Q4_K_M
It used 16.3GB here. To use it comfortably alongside other apps, a Mac with 36GB of memory or more is a good guide (the model then takes about half the memory).
About the model
- Maker
- Mistral
- Size
- 23.6B parameters
- Quantization
- Q4_K_M
- Thinks before answering
- No
- Longest input at once
- 393,216 tokens
- Released (Hugging Face)
- Nov 28, 2025
- Generation speed
- 28.3 tok/s
- Start-up (loading)
- about 1.3 s
- Measured on
- Sep 29, 2026
- Ollama
- 0.34.3
- Hugging Face
- mistralai/Devstral-Small-2-24B-Instruct-2512
- Series
- Mistral
Models of similar memory
- Mistral-Small-3.1-24B-Instruct-2503 (16.3GB of memory, knowledge 72%, reply about 0.2 s)
- Mistral-Small-3.2-24B-Instruct-2506 (16.3GB of memory, knowledge 77%, reply about 0.2 s)
- muse-glimmer:30b (17.6GB of memory, knowledge 85%, reply about 14.3 s)
- LFM2-24B-A2B (14.8GB of memory, knowledge 54%, reply about 0.1 s)
- qwen3.8:27b (18.2GB of memory, knowledge 85%, reply about 6.9 s)
Compare the main models on the text page. The series page lists every model measured.