Model comparison
Mistral Small 4 vs Mistral Large 3
Compare Mistral Small 4 vs Mistral Large 3 API pricing: input/output token costs, cache pricing, context windows, workload estimates, and routing fit.
Mistral Small 4
mistral · mistral/mistral-small-2603
- Input
- $0.15
- Output
- $0.6
- Context
- 256K
Mistral Large 3
mistral · mistral/mistral-large-3
- Input
- $0.5
- Output
- $1.5
- Context
- 262.1K
Quick take
Mistral Small 4 has the lower input price at $0.15 per 1M input tokens. Mistral Small 4 is cheaper for the example blended workload below. Mistral Large 3 has the larger context window at 262.1K tokens.
Choose Mistral Small 4 if...
- Mistral Small 4 is the better default for cost-sensitive traffic and repeated high-volume calls.
Choose Mistral Large 3 if...
- Mistral Large 3 is safer for long documents, repository analysis, and RAG prompts because it has the larger context window.
Example workload cost
Estimates use input tokens plus 20% output tokens. They exclude provider discounts, cache hits, and tool/search surcharges.
| Workload | Mistral Small 4 | Mistral Large 3 | Cheaper |
|---|---|---|---|
| 1M input + 200K output | $0.27 | $0.80 | Mistral Small 4 |
| 10M input + 2M output | $2.70 | $8.00 | Mistral Small 4 |
| 100M input + 20M output | $27.00 | $80.00 | Mistral Small 4 |
Context, output, and capability fit
Mistral Large 3 provides the larger context window. Check max output separately when the task needs long reports, code generation, or full-document rewrites.
- Mistral Small 4 max output
- n/a
- Mistral Large 3 max output
- n/a
- Mistral Small 4 features
- prompt caching, function calling, vision
- Mistral Large 3 features
- prompt caching, function calling, vision
Risk notes for Mistral Small 4
- No major capability risk is flagged in this snapshot, but provider pages should still be verified before production routing.
Risk notes for Mistral Large 3
- No major capability risk is flagged in this snapshot, but provider pages should still be verified before production routing.
Routing tags
budgetcodingreasoningagentsopen-weightenterprisemultimodalraglocal-openfrontier