Mistral Saba
LegacyStandardMistral's multilingual model with strong Arabic and regional language support.
Optimized for Middle Eastern language tasks. Superseded by newer Mistral models.
▼
Mistral Saba is somewhat expensive at $0.20/M tokens input and $0.60/M tokens output. Cached input reads are priced at $0.02/M tokens, making repeated context cheaper. The model supports a 33K context window.
Benchmarks
Specifications
Performance
Providers
Misc
Compare with
Frequently Asked Questions about Mistral Saba▼
When was Mistral Saba released?
▼
Mistral Saba was released on February 17, 2025.
Who created Mistral Saba?
▼
Mistral Saba was created by Mistral.
How much does Mistral Saba cost?
▼
Mistral Saba costs $0.20/M input tokens and $0.60/M output tokens.
What is Mistral Saba API pricing?
▼
The Mistral Saba API is priced at $0.20 per million input tokens and $0.60 per million output tokens. You can access Mistral Saba through idapt.app alongside 200+ other AI models in one workspace.
Is Mistral Saba a reasoning model?
▼
No, Mistral Saba is not a reasoning model. It responds directly without an extended internal thinking step. For tasks requiring deep chain-of-thought reasoning, consider models like Claude, o-series from OpenAI, or Gemini Thinking.
Does Mistral Saba support image or vision input?
▼
No, Mistral Saba does not currently support image or vision input. It is a text-only model.
Does Mistral Saba support audio?
▼
No, Mistral Saba does not support audio input.
What is the context window of Mistral Saba?
▼
Mistral Saba has a context window of 33K tokens, meaning it can process approximately 24,576 words of context at once.
Is Mistral Saba open source?
▼
Mistral Saba is a proprietary model developed by Mistral. The model weights and training data are not publicly available.
Where can I use Mistral Saba?
▼
You can chat with Mistral Saba on idapt.app — no separate API key required. idapt is an AI workspace with 200+ models, agents, computers, and tasks. You can also access Mistral Saba through Mistral's own API or platform.