← Models

Mistral Small 4

Mistral AI · Mistral Small

Mistral's new open-weight small model for efficient long-context assistants and coding support.

Type
language
Context
256K tokens
Max Output
256K tokens
Status
current
Input
$0.15/1M tok
Output
$0.6/1M tok
API Access
Yes
License
Apache 2.0
open-weights mixture-of-experts cost-efficient multilingual coding enterprise
Released March 2026 · Updated April 13, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on April 13, 2026.

Mistral Small 4 is Mistral AI’s latest small-model open-weight release, aimed at teams that want a stronger quality-per-dollar envelope without moving into the cost and infrastructure footprint of large frontier systems. In practical terms, it extends Mistral’s “small but production-usable” lane rather than replacing the large-model tier.

For Signal Lens readers, it matters because it is one of the clearest current examples of an open-weight model that can plausibly sit in real assistant and coding workflows instead of only acting as a local curiosity.

Capabilities

Mistral positions Small 4 as a general long-context model that is still useful for coding, assistant work, and enterprise-style retrieval flows. The model is not framed as a specialist coding release in the way Devstral 2 is, but it is clearly meant to hold up in developer workflows where cost and deployment flexibility matter.

This makes it attractive for high-volume internal copilots, private knowledge assistants, or self-hosted team helpers where a good-enough open model beats a more expensive proprietary default.

Technical Details

Official Mistral docs list Mistral Small 4 as a 26.03 open release with a 256K context window. The same docs publish the model as a 119B-parameter architecture with 6.5B active parameters, which is the most important technical detail because it explains why the model can aim above older “small model” expectations without paying full dense-model cost at inference time.

As with Devstral 2, Mistral’s public docs foreground context-window size rather than a distinct completion ceiling. This profile therefore records maxOutput as the same 256K snapshot limit for consistency with the repo’s comparison model rather than as a separately published provider guarantee.

Pricing & Access

Published Mistral API pricing is:

  • Input: $0.15 per 1M tokens
  • Output: $0.60 per 1M tokens

The model is also available as open weights, which matters more here than it would for many premium proprietary releases. Access choice is part of the value proposition.

Best Use Cases

Use Mistral Small 4 for long-context internal assistants, low-cost code review helpers, multilingual document pipelines, and privacy-sensitive deployments where API spend or data residency is a hard constraint.

It is a better fit for “deploy broadly and cheaply” than for “squeeze every last drop of frontier reasoning performance out of a hard problem.”

Comparisons

  • Mistral Small 3.2 (Mistral AI): Small 4 is the newer higher-capability small-model route with a much larger published context window.
  • Devstral 2 (Mistral AI): Devstral 2 is the better choice for developer-first coding stacks; Small 4 is the broader efficient generalist.
  • Qwen3.5 (Alibaba): Qwen3.5 remains strong for bilingual and APAC-oriented deployments; Small 4 is a strong Western open alternative for global or EU-oriented teams.