Mistral Small 3.2
Mistral Small 3.2 is a chat model from Mistral AI, positioned in the company's Small tier as an incremental update within the Mistral Small 3.x line.
API Pricing
| Provider | Input / 1M | Output / 1M |
|---|---|---|
| $0.094 | $0.250 |
Prices updated daily. Last check: Oct 16, 2026
Compare API pricing for every Mistral model →Mistral Small 3.2 pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Reasoning & Knowledge
- MMLU-Pro68.1%
- GPQA Diamond50.5%
- Humanity's Last Exam4.3%
Coding
- LiveCodeBench27.5%
- SciCode28.6%
Math
- AIME 202527.0%
- AIME32.3%
- MATH-50088.3%
Agentic & Tool Use
- Terminal-Bench Hard6.8%
- Terminal-Bench v2.15.6%
- τ²-bench29.5%
- τ-bench Banking6.2%
Instruction & Long Context
- IFBench33.5%
- Long-Context Reasoning20.3%
Benchmarks measured Oct 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Mistral
- Modalities
- Text
Capabilities
- Open Source
- Yes
Strengths & Limitations
Strengths
- Positioned in Mistral's Small tier, targeting lower serving cost and latency than the company's Medium and Large class models
- Point release within the Mistral Small 3.x line, so prompts and integrations built for the prior 3.x revision generally carry over with minimal changes
- Available from multiple inference providers, which allows price and latency comparison rather than single-vendor lock-in
- Mistral publishes models under a consistent instruct/chat interface, simplifying migration between sizes in the family
- Small-tier sizing makes it a practical candidate for high-volume, repetitive text workloads where a larger model would be over-provisioned
- Provider choice includes options for self-hosting or dedicated deployment where data residency matters
Limitations
- We do not track a confirmed context window for this entry, so maximum input length must be checked with the individual provider
- No independent throughput or time-to-first-token measurements are recorded in our database for this model
- Small-tier models generally trail larger Mistral models on multi-step reasoning, long-form coding, and nuanced instruction following
- Capabilities such as vision input, tool calling, and structured output may differ by host and are not confirmed in our metadata
- As a point update rather than a new generation, gains over the prior 3.x revision are incremental rather than substantial
Key Features
About Mistral Small 3.2
Common Use Cases
Mistral Small 3.2 fits workloads where per-request cost and response latency matter more than maximum reasoning depth: customer support chat, summarization and rewriting pipelines, structured data extraction from short documents, content classification and routing, and retrieval-augmented question answering over modest context. It is also a common choice as the fast tier in a tiered routing setup, handling the majority of straightforward requests while escalating hard or ambiguous cases to a larger Mistral model or another vendor's higher tier. For agentic workflows with long tool chains, deep multi-file code generation, or tasks requiring extended chain-of-thought, a Medium or Large class model or a dedicated reasoning model is typically the better fit.
Frequently Asked Questions
How much does Mistral Small 3.2 cost to run?
Pricing depends on which provider you use and whether you are buying serverless per-token inference, a dedicated endpoint, or GPU capacity to self-host. Input and output tokens are usually billed at different rates, and rates change frequently. See the pricing table on this page for current per-provider figures.
What is Mistral Small 3.2 best used for?
It suits high-volume, latency-sensitive text tasks — support chat, summarization, extraction, classification, and retrieval-augmented Q&A — where a Small-tier model's cost profile is more important than the extra accuracy of a larger model.
How does it differ from Mistral Small 3.1?
Mistral Small 3.2 is a point update within the same 3.x line rather than a new model generation, so changes are incremental and existing prompts and integrations generally transfer. Check Mistral's release notes for the specific behavioral changes in the revision.
What context window does Mistral Small 3.2 support?
Our database does not carry a confirmed context length for this entry, and hosted providers sometimes expose a shorter window than the model's maximum. Verify the supported context with the specific provider you plan to use.
When should I use a larger Mistral model instead?
Move up to a Medium or Large class model when tasks involve multi-step reasoning, long agentic tool chains, complex code generation across many files, or when accuracy on hard prompts justifies the higher per-token cost.