Saba
mistral-sabaMistral Saba is a 24B-parameter language model specifically designed for the Middle East and South Asia, delivering accurate and contextually relevant responses while maintaining efficient performance. Trained on curated regional... Built by Mistral AI.
Prices updated
Input price
$0.20
per 1M tokens · Standard
Output price
$0.60
per 1M tokens · Standard
Input limit
33K
tokens
Output limit
26K
tokens
Input formats
Output formats
Saba price history
2 price records since 9/25/2026
Saba cost calculator
$7.00/month
Standard pricing
Overview
What is Saba?
Saba is a text & reasoning model from Mistral AI. Mistral Saba is a 24B-parameter language model specifically designed for the Middle East and South Asia, delivering accurate and contextually relevant responses while maintaining efficient performance. Trained on curated regional... Its 33K context window and $0.20 input price make it a candidate for cost-sensitive, high-throughput applications.
Benchmarks
Saba benchmarks & speed
How Saba scores on standardized evaluations, and where it lands among every model we track.
6.5
Intelligence Index
Reasoning
GPQA Diamond
42.4%
Graduate-level scientific reasoning · top 93%
HLE
4.3%
Humanity's Last Exam · top 85%
Independent scores from Artificial Analysis and DesignArena · updated 10/5/2026. Higher is better; ranks compare against every model we track with that score.
Rates
Saba pricing
Pricing is based on token usage. Provider-specific caching, batch, regional, and tool charges may affect the final cost.
Input · Standard
$0.20
Per 1M tokens
Cached input · Standard
$0.02
Per 1M tokens
Capabilities
Saba Tools
Tools available when using Saba through supported provider APIs.
Function calling
Supported
Structured outputs
Supported
JSON mode
Supported
Reasoning
Not supported
Built-in web search
Not supported
Log probabilities
Not supported
Deterministic seed
Supported
Parallel tool calls
Not supported
Prompt caching
Supported
Strengths and limitations
Strengths
- Low input cost at $0.20 per million tokens suits high-volume workloads.
- A 33K context window covers most focused application workflows.
- Supports text & reasoning workloads in one model.
Limitations
- Real-world cost still depends on prompt length, response length and provider-specific billing rules.
- The 33K context window is smaller than several long-context alternatives.
- Arena Elo and MMLU-Pro are directional; test accuracy, latency and reliability on your own workload before committing.
Arena Elo, MMLU-Pro and pricing figures are illustrative; validate current vendor terms before purchase.
Same provider