Language modelActive

Mistral Small 4

Mistral AI

Released: March 16, 2026
Data date: September 8, 2026
Predecessor: Mistral Small 3.2

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.

Specifications and access

Specification	Value and source
Weights	Open weightsSource
Access	API, LocalSource
Parameters	119BSource
Active parameters	6.5BSource
Context window	256,000 tokensSource
Knowledge cutoff	November 1, 2024Source
Notes	Current small Mistral API model; sparse MoE with 119B total, 6B active; successor to Mistral Small 3.2Source
License	Apache-2.0Source
Release	2026-03-16Source
Parameters	119 BSource
Active parameters	6.5 B (Release article says 6B active (8B including embedding/output); HF card gives 6.5B active. The source values are not identical.)Source
Architecture	sparse mixture-of-experts, 128 experts with 4 activeSource
Context window	262,144 tokensSource
Input	Text, ImageSource
Output	TextSource
API model ID	mistral-small-2603Source
Reasoning mode	none, highSource
Tool calling	YesSource
Structured output	YesSource

Published benchmarks

Terminal-Bench 2.0

Overall

1.GPT-5.573.2%

2.Claude Opus 4.870.04%

3.Claude Opus 4.768.54%

4.Gemini 3.1 Pro Preview67.42%

4.Gemini 3.5 Flash67.42%

6.GPT-5.3-Codex64.05%

7.Claude Sonnet 4.659.55%

8.Qwen 3.7 Max59.18%

9.Claude Opus 4.558.43%

9.GPT-5.458.43%

11.Kimi K2.657.3%

12.DeepSeek-V4-Pro56.18%

13.GLM-5.153.93%

14.Gemini 3 Flash Preview51.69%

15.GLM-549.44%

16.MiniMax M2.747.19%

17.MiniMax M346.07%

18.GPT-5.4 mini44.94%

18.Qwen 3.6 27B44.94%

20.Grok 4.343.45%

21.MiniMax M2.541.57%

22.Grok 4.20 Reasoning40.45%

22.Kimi K2.540.45%

24.GPT-5.4 nano39.89%

25.Gemma 4 31B39.33%

26.GLM-4.738.2%

27.GPT-537.08%

27.Kimi K2 Thinking37.08%

27.MiniMax M2.137.08%

30.DeepSeek-V3.234.83%

31.Gemini 2.5 Pro30.34%

31.Mistral Medium 3.530.34%

33.GLM-4.628.09%

34.GPT-5 mini26.97%

35.Kimi K2 071125.84%

36.Gemini 3.1 Flash-Lite24.72%

37.DeepSeek-V3.122.47%

38.GPT OSS 120B19.1%

39.Trinity Large Thinking17.98%

40.Mistral Small 416.85%

41.GPT-4.114.61%

42.Mistral Large 38.99%

43.Command A2.25%

higher is betterAs of September 29, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-terminal-bench-2:2.0:overall
Metric: accuracy
Scale: 0-100
Participants: 43
Methodology and context: Source model ID: mistralai/mistral-small-2603. Source model name: Mistral Small 4. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-06-04. Latency in seconds: 407.865. Cost per task in USD: 0.105762. Source ID: vals-ai. Evidence capture ID: vals-terminal-bench-2-2026-09-29-020c4135b825. Source payload hash: 020c4135b82502d8cc0324b9b7d9474024a86da8f4770c78743bd13103cdfa0c. Parser version: vals-astro-v2. Imported at: 2026-09-29

Source: Vals AI

Vibe Code Bench v1.1

Overall

1.Claude Sonnet 5.592.39%

2.Claude Fable 590.35%

3.Claude Opus 5.590.29%

4.Claude Fable 5.190.26%

5.GPT-6 Astra89.59%

6.Claude Opus 588.4%

7.GPT-6 Sol87.82%

8.Grok 4.786.18%

9.Kimi K384.96%

10.DeepSeek-V4.1-Flash84.74%

11.Claude Opus 4.882.73%

12.GPT-6 Luna81.65%

13.Claude Sonnet 581.33%

14.GPT-5.6 Sol80.5%

15.Gemini 3.8 Flash78.65%

16.GLM-5.378.13%

17.GPT-5.6 Luna77.06%

18.Grok 4.676.24%

19.GPT-5.6 Terra74.59%

20.Claude Opus 4.771%

21.Gemini 3.7 Flash70.4%

22.GPT-5.569.85%

23.Grok 4.569%

24.GPT-5.467.42%

25.Qwen 3.8 27B64.85%

26.Qwen 3.8 Max 090264.7%

27.Gemini 3.6 Flash64.01%

28.GLM-5.263.96%

29.GPT-5.3-Codex61.77%

30.Claude Opus 4.657.57%

31.Claude Sonnet 4.651.48%

32.DeepSeek-V4-Pro49.93%

33.Gemini 3.5 Flash48.68%

34.GPT-5.4 mini47.97%

35.Qwen 3.7 Max47.67%

36.MiniMax M347.57%

37.Kimi K2.7 Code47.21%

38.Qwen 3.7 Plus46.39%

39.MiMo-V2.542.17%

40.Kimi K2.637.89%

41.Gemini 3.5 Flash-Lite37.16%

42.MiMo-V2.5-Pro34.11%

43.Gemini 3.1 Pro Preview32.03%

44.GLM-5.131.46%

45.GLM-5.3-Flash30.76%

46.GPT-5.4 nano26.1%

47.GLM-523.36%

48.Gemini 3 Flash Preview20.2%

49.GPT-520.09%

50.Grok 4.319.4%

51.Inkling19.21%

52.Inkling Small19.05%

53.Kimi K2.517.54%

54.MiniMax M2.514.85%

55.GPT-5 mini14.17%

56.Qwen 3.6 27B11.94%

57.MiniMax M2.711.93%

58.Nemotron 3 Ultra7.64%

59.Grok 4.20 Reasoning4.06%

60.Mercury 2.53.68%

61.GLM-4.63.09%

62.Ling 3.0 Flash2.91%

63.Mistral Medium 3.52.89%

64.Gemini 2.5 Pro0.4%

65.Gemini 3.1 Flash-Lite0%

65.Mistral Small 40%

65.Nemotron 3.5 Lightning0%

higher is betterAs of September 29, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-vibe-code-bench:1.1:overall
Metric: accuracy
Scale: 0-100
Participants: 67
Methodology and context: Source model ID: mistralai/mistral-small-2603. Source model name: Mistral Small 4. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-27. Harness: OpenHands. Latency in seconds: 1814.681. Cost per task in USD: 6.473001. Source ID: vals-ai. Evidence capture ID: vals-vibe-code-2026-09-29-a944ac42bd14. Source payload hash: a944ac42bd147ecb4e54c905d243c5ff8463f0d04b735f1a8627b369fd351391. Parser version: vals-astro-v2. Imported at: 2026-09-29

Source: Vals AI

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

AA-LCR (Show measurement, test conditions, and source)

Source value: 0.72
Score: 0.72
Metric: ratio
Unit: ratio
Direction: Higher is better
Source type: vendor-reported
Evaluator: Mistral AI release post citing Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. RawFraction [0,1]; Artificial Analysis result as quoted by Mistral; exact snapshot/configuration not stated
Context: Vendor release text cites the external evaluator but does not provide the full harness configuration.

Source Mistral AI release post citing Artificial Analysis

Artificial Analysis Intelligence Index (Show measurement, test conditions, and source)

Source value: 11.4511294141469
Score: 11.4511294141469
Metric: points
Unit: points
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown

Source Artificial Analysis

AA-Briefcase Elo (Show measurement, test conditions, and source)

Source value: 255.96
Score: 255.96
Metric: elo
Unit: elo
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown

Source Artificial Analysis

AA-Omniscience Index (Show measurement, test conditions, and source)

Source value: -30.4
Score: -30.4
Metric: points
Unit: points
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown

Source Artificial Analysis

Output speed (Show measurement, test conditions, and source)

Source value: 166.852525395502
Score: 166.852525395502
Metric: tokens/s
Unit: tokens/s
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown
Context: Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: first-party API.

Source Artificial Analysis

Time per Intelligence Index Task (Show measurement, test conditions, and source)

Source value: 1.6497210044870998
Score: 1.6497210044870998
Metric: minutes
Unit: minutes
Direction: Lower is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown
Context: Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: first-party API.

Source Artificial Analysis

Latency reasoning time (Show measurement, test conditions, and source)

Source value: 11.986633077679002
Score: 11.986633077679002
Metric: seconds
Unit: seconds
Direction: Lower is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown
Context: Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: first-party API.

Source Artificial Analysis

Latency input time (Show measurement, test conditions, and source)

Source value: 0.733454756500009
Score: 0.733454756500009
Metric: seconds
Unit: seconds
Direction: Lower is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown
Context: Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: first-party API.

Source Artificial Analysis

Answer generation time (end-to-end timing breakdown) (Show measurement, test conditions, and source)

Source value: 2.9966582694197506
Score: 2.9966582694197506
Metric: seconds
Unit: seconds
Direction: Lower is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown
Context: Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: first-party API.

Source Artificial Analysis

End-to-end reasoning time component (Show measurement, test conditions, and source)

Source value: 11.986633077679002
Score: 11.986633077679002
Metric: seconds
Unit: seconds
Direction: Lower is better
Source type: independent-evaluator
Evaluator: Artificial Analysis
Status: active
Retrieved at: 2026-09-08
Methodology: Independent evaluation. unknown
Context: Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: first-party API.

Source Artificial Analysis

Recorded measurement details

Terminal-Bench 2.0: Overall (Show measurement, test conditions, and source)

Source value: 16.854
Score: 16.854
Metric: accuracy
Unit: %
Benchmark version: 2.0
Category: agentic
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-terminal-bench-2:2.0:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-29
Methodology: Source model ID: mistralai/mistral-small-2603. Source model name: Mistral Small 4. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-06-04. Latency in seconds: 407.865. Cost per task in USD: 0.105762. Source ID: vals-ai. Evidence capture ID: vals-terminal-bench-2-2026-09-29-020c4135b825. Source payload hash: 020c4135b82502d8cc0324b9b7d9474024a86da8f4770c78743bd13103cdfa0c. Parser version: vals-astro-v2. Imported at: 2026-09-29

Source Vals AI

Vibe Code Bench v1.1: Overall (Show measurement, test conditions, and source)

Source value: 0
Score: 0
Metric: accuracy
Unit: %
Benchmark version: 1.1
Category: coding
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-vibe-code-bench:1.1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-29
Methodology: Source model ID: mistralai/mistral-small-2603. Source model name: Mistral Small 4. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-27. Harness: OpenHands. Latency in seconds: 1814.681. Cost per task in USD: 6.473001. Source ID: vals-ai. Evidence capture ID: vals-vibe-code-2026-09-29-a944ac42bd14. Source payload hash: a944ac42bd147ecb4e54c905d243c5ff8463f0d04b735f1a8627b369fd351391. Parser version: vals-astro-v2. Imported at: 2026-09-29
Comparison methodology: Harness: OpenHands

Source Vals AI

Page 1 of 2

Predecessor and successor

Sources and data date

Type	Evidence and data date
Speed	Mistral. Output speed after the first streaming chunk, not total response time. Retrieved September 8, 2026. Source: Artificial Analysis · Artificial Analysis (retrieved September 8, 2026)Speed measurement methodology
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · API pricing
Input	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Model metadata (retrieved September 8, 2026) · Mistral Small 4 model card · Editorial description reviewed October 2, 2026
Output	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Model metadata (retrieved September 8, 2026) · Mistral Small 4 model card · Editorial description reviewed October 2, 2026
Research status	Research date September 8, 2026. 4 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	knowledge-cutoff Original wording Neither the official docs, release post nor model card publishes a cutoff date.
Not found in the inspected sources	training-tokens Original wording No training-token count found in the checked primary sources.
Not found in the inspected sources	max-output-tokens Original wording Checked model page does not state a model-specific maximum output.
Not found in the inspected sources	profile-benchmarks Original wording The checked card/release has narrative benchmark claims but no additional exact score table.
Additional source	Model metadata (retrieved September 8, 2026) · Mistral Small 4 announcement · Editorial description reviewed October 2, 2026
Additional source	Knowledge cutoff data
Additional source	Vals AI (retrieved September 29, 2026)
Additional source	Vals AI (retrieved September 29, 2026)
Additional source	Model metadata (retrieved September 8, 2026)
Additional source	Mistral model overview · Editorial description reviewed October 2, 2026
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Creative Commons Attribution 4.0

Mistral Small 4

Specifications and access

Published benchmarks

Terminal-Bench 2.0

Vibe Code Bench v1.1

Measurements without matching peer values

Recorded measurement details

Our tests

Pelican on a bicycle

Popularity

Published prices

Predecessor and successor

Sources and data date