Mistral Small 24B Instruct 2501
PermissiveMistral AI · Mistral · Released 2025-01
24BText
A dense 24B tuned to fit on a single high-end GPU, aimed squarely at teams that want Apache-licensed weights without MoE serving complexity.
Strengths
- +Fully permissive Apache 2.0 license
- +Fits on a single 24-32GB GPU at reasonable quantization
- +Competitive with larger models on knowledge tasks
Limitations
- -Smaller community/tooling footprint than Llama or Qwen at this size
- -32K context is modest versus 2025-era peers
License
Apache 2.0
use commercially with attribution niceties
Hardware
wants 24-48GB of VRAM - a 3090/4090-class card or better
Links
Stats
mistralai/Mistral-Small-24B-Instruct-2501— downloads·— likes
via Hugging Face