Mistral: Saba
Mistral Saba from MistralAI is designed for handling text and file inputs with a context length of 32,768 tokens. It supports integrating external tools but lacks reasoning capabilities or structured output formats, making it more suitable for tasks that require interaction with other systems. The model has a blended benchmark score of 9.1 across independent tests; while its performance is strong, the cost at $0.2 per million input tokens and $0.6 per million output tokens could be a consideration for budget-conscious users needing robust tool integration.
Benchmark results
Independent, published benchmarks. Blended score 9.1 across 1 benchmark, last refreshed 2026-07-21. How scoring works →
| Benchmark | Measures | Score |
|---|---|---|
| AI Index | broad capability composite | 10.6 |
- Model ID
- mistralai/mistral-saba
- Vendor
- mistralai
- Released
- February 2025
- Tokenizer
- Mistral
- Input Modalities
- text, file
- Output Modalities
- text
- Max Output
- default
- Tool Calling
- ✓ supported
- Structured Output
- ✓ supported
- Reasoning Mode
- not supported
- Vision
- text only
- Audio
- no
- Moderated
- no
What it costs in practice
Computed from the current $0.20/M input and $0.60/M output rates. Run your own numbers →
| Job | Tokens | Cost |
|---|---|---|
| Summarize a 50-page report | 30k in / 1.5k out | under $0.01 |
| Classify 1,000 customer emails | 500k in / 50k out | $0.13 |
| A month of a busy support chatbot | 5M in / 2M out | $2.20 |
Similar models
Mistral: Mixtral 8x22B Instruct
Mistral: Mistral Small 3.1 24B
Mistral: Voxtral Small 24B 2507
Mistral: Mistral Nemo
Mistral Large
Mistral Large 2407
Quick answers
- How much does Mistral: Saba cost?
- $0.20 per million input tokens and $0.60 per million output tokens.
- What is Mistral: Saba's context window?
- 32,768 tokens, roughly 49 pages of text in a single request.
- Does Mistral: Saba support tool calling?
- It supports tool calling, structured output.
- Can Mistral: Saba process images?
- No, it is text-only on the input side.