Skip to main content
Language modelActive

Mistral Small 4

Mistral AI

Released
March 16, 2026
Data date
September 8, 2026
Benchmark-Meta-Score
0 / 100
Standard error ± 0
Speed
166.9 tokens/s
Output throughput
Price
$0.15 · $0.6
Input · Output / 1M tokens
Input
Text, Image
Output
Text

What is distinctive about Mistral Small 4 is the combination of instruction following, reasoning, image understanding, and agentic coding. Its sparse mixture-of-experts design has 119 billion total parameters, 128 experts, and four active experts per token. It accepts text and images with up to 256,000 tokens of context; reasoning ranges from none to high.

Mistral brings together directions represented by Mistral Small 3.2, Magistral Small 1.2, Pixtral 12B, and Devstral Small. The no-reasoning mode remains available.

Position
Rank 82 of 91
Index score
0 / 100
1.Claude Fable 5.193 / 100
1.Claude Opus 5.594.3 / 100
1.Claude Sonnet 5.591.5 / 100
2.GPT-6 Astra90.5 / 100
3.Claude Opus 587.6 / 100
4.GPT-6.1 Sol88.6 / 100
4.Muse Wonder Full86.9 / 100
5.Claude Fable 585 / 100
7.Muse Wonder80 / 100
10.Gemini 3.7 Flash71.3 / 100
10.Gemini 3.8 Flash70.2 / 100
10.GPT-5.6 Sol71.4 / 100
10.GPT-6 Sol70.4 / 100
10.Muse Spark 1.366.1 / 100
11.Claude Haiku 5.559.3 / 100
13.Kimi K362 / 100
15.GPT-5.554.7 / 100
16.Claude Opus 4.852 / 100
16.Grok 4.655.7 / 100
16.Grok 4.754.6 / 100
17.Qwen 3.8 Max 090249.8 / 100
18.Claude Opus 4.744.9 / 100
18.GLM-5.350.5 / 100
18.GPT-5.6 Terra47.9 / 100
20.DeepSeek-V4.1-Flash42.4 / 100
20.Gemini 3.6 Flash44.4 / 100
21.Gemini 3.5 Flash45.9 / 100
22.Grok 4.537.9 / 100
23.Mistral Large 432.4 / 100
24.GLM-5.3-Flash38 / 100
24.GPT-6 Luna38.6 / 100
28.Claude Sonnet 531.5 / 100
28.GPT-5.6 Luna33.1 / 100
29.Gemini 3.1 Pro Preview28.7 / 100
32.Claude Opus 4.617.9 / 100
33.DeepSeek-V4-Flash15.1 / 100
34.Claude Sonnet 4.620.1 / 100
34.DeepSeek-V4-Pro19.6 / 100
34.GLM-5.224.1 / 100
35.Qwen 3.8 27B14.8 / 100
36.MiniMax M316.9 / 100
36.Qwen 3.7 Max16.8 / 100
37.Gemini 3 Flash Preview10.6 / 100
37.GPT-5.415.4 / 100
40.Kimi K2.613.2 / 100
42.Kimi K2.7 Code9.4 / 100
43.GPT-57.5 / 100
44.GLM-5.18.9 / 100
44.Qwen 3.7 Plus8.2 / 100
45.Claude Opus 4.55.2 / 100
47.Inkling4.6 / 100
47.Inkling Small4.9 / 100
47.MiMo-V2.5-Pro4.5 / 100
49.Gemini 3.5 Flash-Lite4.8 / 100
49.o34.3 / 100
51.GPT-5.4 mini3.7 / 100
54.GLM-52.5 / 100
54.Grok 4.32.2 / 100
55.Kimi K2.52.2 / 100
55.Qwen 3.6 27B1.8 / 100
57.Claude Sonnet 4.50.9 / 100
57.Gemma 4 31B0.3 / 100
57.MiMo-V2.51.4 / 100
58.GPT-5 mini1.4 / 100
60.MiniMax M2.71.1 / 100
60.o4-mini0.7 / 100
62.Gemini 2.5 Pro0.4 / 100
62.GPT-5.4 nano0.7 / 100
63.MiniMax M2.50.6 / 100
64.Gemini 3.1 Flash-Lite0.4 / 100
66.MiniMax M2.10.2 / 100
67.Ling 3.0 Flash0.1 / 100
67.o10.1 / 100
69.Kimi K2 Thinking0.1 / 100
72.GLM-4.50 / 100
73.DeepSeek-R10 / 100
73.Mistral Medium 3.50 / 100
74.GPT OSS 120B0 / 100
75.GPT-4.10 / 100
78.GPT-4.1 mini0 / 100
78.GPT-5 nano0 / 100
78.o3-mini0 / 100
80.Trinity Large Thinking0 / 100
81.DeepSeek-V3 03240 / 100
82.GPT OSS 20B0 / 100
82.Mistral Small 40 / 100
83.Mistral Large 30 / 100
86.DeepSeek-V30 / 100
88.Llama 4 Maverick0 / 100
90.Command A0 / 100
90.Llama 4 Scout0 / 100
Methodology and source

The LLM comparison hub's selected-panel index averages rank percentiles from 65 documented comparison cohorts. Benchmark meta-scores are shown in the benchmark section.