What Is Mistral Small 4?

Mistral Small 4 is Mistral AI’s efficient hybrid model — 119B parameters, 6.5B active, 256K context. Released March 16, 2026.

Last verified: March 29, 2026

Key Specs

Feature	Detail
Developer	Mistral AI (Paris, France)
Release date	March 16, 2026
Total parameters	119B
Active parameters	6.5B per token
Architecture	Hybrid (Mixture-of-Experts)
Context window	256K tokens
Positioning	Sovereign, deployable, high-performance

Why It Matters

Mistral Small 4 sits at a unique intersection:

1. Efficient — Only 6.5B active parameters means lower compute costs
2. Capable — 119B total parameters provide strong reasoning depth
3. Deployable — Runs on moderate infrastructure for on-premises deployment
4. Sovereign — European companies can keep data in-region for GDPR compliance
5. Long context — 256K tokens handles large codebases and documents

Competitive Landscape

Model	Total Params	Active Params	Context
Mistral Small 4	119B	6.5B	256K
GPT-5.4 Mini	Undisclosed	Undisclosed	128K
Claude 4.5 Haiku	Undisclosed	Undisclosed	200K
Qwen 3.5 Small	22B	22B (dense)	128K

Mistral’s MoE architecture gives it the best efficiency ratio in this tier — more total knowledge with lower per-token compute.

Who Should Use It

• European companies needing data sovereignty
• Teams wanting on-premises AI without cloud dependency
• Applications requiring long context (legal documents, codebases)
• Cost-sensitive deployments where per-token efficiency matters