English Français العربية

Mistral Small 4

Mistral AI Language models March 2026

Mistral Small 4 is a model from Mistral AI, listed under Language models. It was released in March 2026. It accepts up to 256k tokens per request.

Add to favourites Documentation Official pricing
Headline price
No data
Context window
256k tokens

Overview

Combines text, reasoning and code in one model under a true open-source licence. Apache-2.0 imposes no user-count ceiling or field-of-use limit, unlike most 'open' model licences.

  • Commercial self-hosting
  • Unrestricted licensing
  • On-premise deployment

Current prices

No public API price
This model is distributed as open weights. Running cost depends on your own hardware.

Price history

No price change recorded for this model yet.

How it compares

Context windowTokens per request. Further right holds more. Logarithmic scale: each step is ten times the last.
10k30k100k300k1MGPT-6 Astra: 1.05MGPT-6.1 Sol: 1.05MGPT-6 Luna: 1.05MGPT-6 Sol: 1.05MGPT-5.6 Sol: 1.05MGPT-5.6 Terra: 1.05MGPT-5.6 Luna: 1.05MGPT-6 Astra Pro: 1.05MGPT-6 Luna Pro: 1.05MGPT-6 Sol Pro: 1.05MGPT-6.1 Sol Pro: 1.05MGPT-6.1 Terra: 1.05MGPT-6.1 Terra Pro: 1.05MGPT-5.4: 1.05MGPT-5.4 Pro: 1.05MGPT-5.5: 1.05MGPT-5.5 Pro: 1.05MGemini 4 Argon: 1.05MGemini 3.8 Flash: 1.05MGemini Omni Flash: 1.05MMuse Wonder: 1.05MGemini 3.5 Flash-Lite: 1.05MGemini 3.7 Flash: 1.05MDeepSeek V4 Flash 0731: 1.05MMuse Spark 1.3: 1.05MKimi K3.1 Pro: 1.05MQwen3.8 2.4T A95B: 1.05MGemini 3.1 Pro: 1.05MDeepSeek V4 Flash Vision Exp: 1.05MGemini 3.6 Flash: 1.05MGemini 3.1 Flash-Lite: 1.05MGemini 3 Flash: 1.05MGemini 2.5 Pro: 1.05MGemini 2.5 Flash: 1.05MGemini 2.5 Flash-Lite: 1.05MMuse Spark 1.1: 1.05MMuse Spark 1.2: 1.05MGPT-4.1: 1.05MGPT-4.1 Mini: 1.05MGPT-4.1 nano: 1.05MClaude Fable 5.1: 1MClaude Opus 5.5: 1MClaude Sonnet 5.5: 1MClaude Haiku 5.5: 1MGrok 4.3: 1MDeepSeek V4 Pro: 1MDeepSeek V4.1 Flash: 1MKimi K3: 1MClaude Mythos 5.1: 1MGrok 4.20 Reasoning: 1MGrok 4.20 Non-Reasoning: 1MGrok 4.20 Multi-Agent: 1MZ.ai GLM 5.3 (on Mistral): 1MQwen3.8 27B: 1MQwen3.8 Max Prime: 1MMistral Large 4: 1MClaude Fable 5: 1MClaude Mythos 5: 1MClaude Opus 5: 1MClaude Sonnet 5: 1MClaude Opus 4.8: 1MClaude Opus 4.7: 1MClaude Opus 4.6: 1MClaude Sonnet 4.6: 1MGrok 4.7: 500kGrok 4.6: 500kGrok 4.5: 500kGPT-5.6 Cyber: 400kChat Latest: 400kGPT-5: 400kGPT-5 Mini: 400kGPT-5 nano: 400kGPT-5 Pro: 400kGPT-5.2: 400kGPT-5.2 Pro: 400kGPT-5.4 Mini: 400kGPT-5-Codex: 400kGPT-5.1: 400kGPT-5.1-Codex: 400kGPT-5.1-Codex Mini: 400kGPT-5.1-Codex-Max: 400kGPT-5.2-Codex: 400kGPT-5.3-Codex: 400kGPT-5.4 nano: 400kGrok Build 0.1: 256kMistral Medium 3.5: 256kMinistral 3 14B: 256kMinistral 3 8B: 256kMinistral 3 3B: 256kCommand A: 256kCommand A Reasoning: 256kNorth Mini Code: 256kKimi K2.7 Code: 256kMistral Large 3: 256kKimi K2.6: 256kClaude Opus 4.5: 200kClaude Haiku 4.5: 200ko3: 200ko3-pro: 200kClaude Sonnet 4.5: 200kcodex-mini-latest: 200ko1: 200ko1-pro: 200ko3-deep-research: 200ko3-mini: 200ko4-mini: 200ko4-mini-deep-research: 200kgpt-oss-120b: 131kgpt-oss-20b: 131kCommand A+: 128kCodestral: 128kAya Expanse 32B: 128kCommand A Vision: 128kCommand R7B: 128kCommand R (08-2024): 128kCommand R+ (08-2024): 128kChatGPT-4o: 128kGPT-4 Turbo: 128kGPT-4 Turbo Preview: 128kGPT-4.5 Preview: 128kGPT-4o: 128kGPT-4o Mini: 128kGPT-4o Mini Search Preview: 128kGPT-4o Search Preview: 128kGPT-5 Chat: 128kGPT-5.1 Chat: 128kGPT-5.2 Chat: 128kGPT-5.3 Chat: 128ko1 Preview: 128ko1-mini: 128kCommand R: 128kCommand R+: 128kVoxtral Small: 32kGPT-3.5 Turbo: 16kbabbage-002: 16kdavinci-002: 16kAya Vision 32B: 16kcomputer-use-preview: 8kGPT-4: 8kTiny Aya Earth: 8kTiny Aya Global: 8kMistral Small 4: 256k256k
Show data
ModelValue
GPT-6 Astra1.05M
GPT-6.1 Sol1.05M
GPT-6 Luna1.05M
GPT-6 Sol1.05M
GPT-5.6 Sol1.05M
GPT-5.6 Terra1.05M
GPT-5.6 Luna1.05M
GPT-6 Astra Pro1.05M
GPT-6 Luna Pro1.05M
GPT-6 Sol Pro1.05M
GPT-6.1 Sol Pro1.05M
GPT-6.1 Terra1.05M
GPT-6.1 Terra Pro1.05M
GPT-5.41.05M
GPT-5.4 Pro1.05M
GPT-5.51.05M
GPT-5.5 Pro1.05M
Gemini 4 Argon1.05M
Gemini 3.8 Flash1.05M
Gemini Omni Flash1.05M
Muse Wonder1.05M
Gemini 3.5 Flash-Lite1.05M
Gemini 3.7 Flash1.05M
DeepSeek V4 Flash 07311.05M
Muse Spark 1.31.05M
Kimi K3.1 Pro1.05M
Qwen3.8 2.4T A95B1.05M
Gemini 3.1 Pro1.05M
DeepSeek V4 Flash Vision Exp1.05M
Gemini 3.6 Flash1.05M
Gemini 3.1 Flash-Lite1.05M
Gemini 3 Flash1.05M
Gemini 2.5 Pro1.05M
Gemini 2.5 Flash1.05M
Gemini 2.5 Flash-Lite1.05M
Muse Spark 1.11.05M
Muse Spark 1.21.05M
GPT-4.11.05M
GPT-4.1 Mini1.05M
GPT-4.1 nano1.05M
Claude Fable 5.11M
Claude Opus 5.51M
Claude Sonnet 5.51M
Claude Haiku 5.51M
Grok 4.31M
DeepSeek V4 Pro1M
DeepSeek V4.1 Flash1M
Kimi K31M
Claude Mythos 5.11M
Grok 4.20 Reasoning1M
Grok 4.20 Non-Reasoning1M
Grok 4.20 Multi-Agent1M
Z.ai GLM 5.3 (on Mistral)1M
Qwen3.8 27B1M
Qwen3.8 Max Prime1M
Mistral Large 41M
Claude Fable 51M
Claude Mythos 51M
Claude Opus 51M
Claude Sonnet 51M
Claude Opus 4.81M
Claude Opus 4.71M
Claude Opus 4.61M
Claude Sonnet 4.61M
Grok 4.7500k
Grok 4.6500k
Grok 4.5500k
GPT-5.6 Cyber400k
Chat Latest400k
GPT-5400k
GPT-5 Mini400k
GPT-5 nano400k
GPT-5 Pro400k
GPT-5.2400k
GPT-5.2 Pro400k
GPT-5.4 Mini400k
GPT-5-Codex400k
GPT-5.1400k
GPT-5.1-Codex400k
GPT-5.1-Codex Mini400k
GPT-5.1-Codex-Max400k
GPT-5.2-Codex400k
GPT-5.3-Codex400k
GPT-5.4 nano400k
Grok Build 0.1256k
Mistral Medium 3.5256k
Ministral 3 14B256k
Ministral 3 8B256k
Ministral 3 3B256k
Command A256k
Command A Reasoning256k
North Mini Code256k
Kimi K2.7 Code256k
Mistral Large 3256k
Kimi K2.6256k
Claude Opus 4.5200k
Claude Haiku 4.5200k
o3200k
o3-pro200k
Claude Sonnet 4.5200k
codex-mini-latest200k
o1200k
o1-pro200k
o3-deep-research200k
o3-mini200k
o4-mini200k
o4-mini-deep-research200k
gpt-oss-120b131k
gpt-oss-20b131k
Command A+128k
Codestral128k
Aya Expanse 32B128k
Command A Vision128k
Command R7B128k
Command R (08-2024)128k
Command R+ (08-2024)128k
ChatGPT-4o128k
GPT-4 Turbo128k
GPT-4 Turbo Preview128k
GPT-4.5 Preview128k
GPT-4o128k
GPT-4o Mini128k
GPT-4o Mini Search Preview128k
GPT-4o Search Preview128k
GPT-5 Chat128k
GPT-5.1 Chat128k
GPT-5.2 Chat128k
GPT-5.3 Chat128k
o1 Preview128k
o1-mini128k
Command R128k
Command R+128k
Voxtral Small32k
GPT-3.5 Turbo16k
babbage-00216k
davinci-00216k
Aya Vision 32B16k
computer-use-preview8k
GPT-48k
Tiny Aya Earth8k
Tiny Aya Global8k
Mistral Small 4256k

Benchmarks

Benchmarks Score Setting tested
Intelligence Index Artificial Analysis's composite score across about ten independent evaluations (reasoning, knowledge, coding, agentic work, long context), run the same way for every model. The strongest models score in the 50s. 11.3 Reasoning
Terminal-Bench Hard Hard command-line tasks (building software, configuring systems, fixing environments) completed autonomously in a real terminal. A direct test of agentic coding. 17.4% Reasoning
SciCode Writing code to solve research-level problems in physics, chemistry, biology and maths, checked by tests. 38.8% Reasoning
GPQA Diamond Graduate-level questions in biology, physics and chemistry, written to be resistant to web search. PhD holders in the relevant field score around 65%, so scores above that indicate genuine domain reasoning rather than recall. 76.9% Reasoning
Humanity's Last Exam About 2,500 expert-written questions at the frontier of human knowledge across maths, science and the humanities. Even the best models answer well under two thirds. 9.9% Reasoning
IFBench (instruction following) Precise, verifiable formatting and content constraints the model has not seen in training. Measures how faithfully it follows instructions. 48.2% Reasoning
τ²-Bench (agentic tool use) Simulated customer-support conversations where the model must use tools and follow policy while a simulated user acts on their side. Measures reliable multi-turn tool use. 41.2% Reasoning
MMMU-Pro (visual reasoning) College-level questions that require reading charts, diagrams and figures, with distractors that defeat text-only guessing. 56.8% Reasoning
Long-context reasoning (AA-LCR) Questions that require reasoning across several long documents at once (around 100k tokens), not just finding one passage. 49.7% Reasoning

Independent evaluations by Artificial Analysis, run the same way for every model, at the reasoning setting shown. Percentages unless marked as an index. Source

Alternatives

Same category, closest price, other providers first.

Grok 4.20 Reasoning xAI

$1.56 / 1M tokens

Compare
Grok 4.20 Non-Reasoning xAI

$1.56 / 1M tokens

Compare
Grok 4.20 Multi-Agent xAI

$1.56 / 1M tokens

Compare
Tiny Aya Global Cohere

No data

Compare

More from Mistral AI

Mistral AI
# Model Input Output Context Released Capabilities Price change
1
Mistral Medium 3.5 Mistral AI
No data No data 256k April 2026
2
Ministral 3 14B Mistral AI
No data No data 256k December 2025
3
Ministral 3 8B Mistral AI
No data No data 256k December 2025
4
Ministral 3 3B Mistral AI
No data No data 256k December 2025
5
No data No data 1M Aug 18, 2026
6
Codestral Mistral AI
No data No data 128k August 2025
Spotted something wrong? Report it

Tell us what is wrong on this page. We check every report against the provider's own documentation before changing anything.

Frequently asked questions

What is the context window of Mistral Small 4?

Mistral Small 4 accepts up to 256k tokens per request.

Is Mistral Small 4 still available?

Yes. Mistral Small 4 is generally available through the Mistral AI API.

When was Mistral Small 4 released?

Mistral Small 4 was released in March 2026.

What is the API model ID for Mistral Small 4?

Use mistral-small-2603 when calling the Mistral AI API.

What are the alternatives to Mistral Small 4?

Models in the same category at a similar price include Grok 4.20 Reasoning (xAI), Grok 4.20 Non-Reasoning (xAI), Grok 4.20 Multi-Agent (xAI) and Tiny Aya Global (Cohere).