Model Details
Mistral Small 3 24B Base
Quick answer: Mistral Small 3 24B Base is the pretrained base (non-instruction) model of Mistral Small 3, scoring 91.3% ARC-C, 80.7% GSM8K, and 54.4% MMLU-Pro. Apache 2.0.
At a Glance
Where Mistral Small 3 Base leads
- 91.3% ARC-C — strong reasoning at base level
- Apache 2.0 — fully open
- Foundation for Mistral Small 3 instruction fine-tuning
Where it lags
- Not instruction-tuned: use Mistral Small 3 Instruct for chat/tasks
- 54.4% MMLU-Pro — lower than instruct variant (66.3%)
Best for: Custom fine-tuning on the Mistral Small 3 architecture; base-model research.
What Mistral Small 3 24B Base Is
The pretrained base model underpinning the Mistral Small 3 24B instruction family. For direct deployment: use Mistral Small 3 24B Instruct instead.
Specifications
| Field | Value |
|---|---|
| Organization | Mistral AI |
| License | Apache 2.0 |
| HuggingFace | mistralai/Mistral-Small-3-24B-Base |
| Release date | January 30, 2025 |
| Parameters | 24B |
| Modality | Text only |
Pricing
Open weights under Apache 2.0 — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| ARC-C | 91.3% | Benchgen evaluation | 2025-01 |
| GSM8K | 80.7% | Benchgen evaluation | 2025-01 |
| MMLU-Pro | 54.4% | Benchgen evaluation | 2025-01 |
Mistral Small 3 Base vs Alternatives
| Model | ARC-C | MMLU-Pro | Type |
|---|---|---|---|
| Mistral Small 3 Base | 91.3% | 54.4% | Base |
| Mistral Small 3 Instruct | — | 66.3% | Instruct |
Always prefer Mistral Small 3 Instruct for direct task deployment; use Base only for fine-tuning.
Related Models & Benchmarks
- Compare: Mistral Small 3 24B Instruct · Mistral Small 3.1 24B
- Benchmarks: ARC-C · GSM8K · MMLU-Pro
Frequently Asked Questions
What is Mistral Small 3 24B Base?
The pretrained base of Mistral Small 3, scoring 91.3% ARC-C, 80.7% GSM8K. Apache 2.0 — use Instruct variant for direct deployment.Specs from Mistral AI's Mistral Small 3 24B Base release (January 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.