Saba
Mistral· released 17 Feb 2025
mistralai/mistral-saba- Input / 1M tokens
- $0.2
- Output / 1M tokens
- $0.6
Longest answer: 26K tokens
About
Saba is a Mistral language model in the mistralai family, released on 2025-02-17. Mistral describes it as a 24B-parameter model designed specifically for the Middle East and South Asia, with responses intended to remain accurate and contextually relevant while maintaining efficient performance. The summary notes training on curated regional material. Saba accepts text and file inputs and produces text outputs. It supports tools, has a 32,768-token context window, and allows a maximum output of 26,214 tokens. Pricing is 0.2 USD per 1M tokens for input and 0.6 USD per 1M tokens for output. Mistral is listed as both maker and lab. The official site is its OpenRouter page, and the model ID is mistralai/mistral-saba. Its modality is text, with files included alongside text as supported inputs. The regional focus distinguishes its intended language-model use from general-purpose specifications.
Who it is for
Saba suits builders targeting Middle East and South Asia use cases that need region-focused language responses. It also fits text or file workflows requiring tools and a 32,768-token context window.
What is good
- Designed for Middle East and South Asia
- 24B-parameter language model
- Supports text and file inputs
- Tools available
- 26,214-token maximum output
What to know first
- Input costs 0.2 USD per 1M tokens
- Output costs 0.6 USD per 1M tokens
- 32,768-token context window
Verdict
Saba is a region-focused Mistral model with text and file inputs, tool support, and a 32,768-token context window. Its listed pricing is 0.2 USD per 1M input tokens and 0.6 USD per 1M output tokens.
Details
- Lab
- Mistralopenrouter.ai · 3 Oct 2026
- Context
- 33Kopenrouter.ai · 3 Oct 2026
- Input price
- $0.2 / 1Mopenrouter.ai · 3 Oct 2026
- Output price
- $0.6 / 1Mopenrouter.ai · 3 Oct 2026
- Max output
- 26,214 tokensopenrouter.ai · 3 Oct 2026
- Inputs
- text, fileopenrouter.ai · 3 Oct 2026
- Open weights
- Noopenrouter.ai · 3 Oct 2026
More longest-context models
See the list
Listed on Inferse
- Longest-Context Models in 2026466 listed
- Newest Models in 2026466 listed
- Cheapest Language Models in 2026437 listed
- Models with Tool Calling in 2026395 listed
- Mistral Models: Pricing per 1M Tokens and Context (2026)23 listed
Sources
- openrouter.ai/mistralai/mistral-saba· checked 3 Oct 2026