We're releasing Mistral Small 3! - 24B params, 81% MMLU - Latency optimized: 150 tokens/s - Competitive with Llama-3.3 70B, Qwen-2.5 32B, GPT4o-mini - Apache 2.0 mistral.ai/news/mistral... - ThreadSky

dlsq.bsky.social • 34 days ago

We're releasing Mistral Small 3!
- 24B params, 81% MMLU
- Latency optimized: 150 tokens/s
- Competitive with Llama-3.3 70B, Qwen-2.5 32B, GPT4o-mini
- Apache 2.0
https://mistral.ai/news/mistral-small-3/

Comments

chanclita.bsky.social•34 days ago

owo

sebastiandeutsch.bsky.social•28 days ago

Congratulations! All your models kick ass!

maxkannen.bsky.social•34 days ago

Are you planning on releasing even smaller models in the near future? 24B is still to big to experiment with on personal hardware.

dlsq.bsky.social•34 days ago

Have you tried our 8B model?
https://huggingface.co/mistralai/Ministral-8B-Instruct-2410

maxkannen.bsky.social•32 days ago

Actually, I forgot that this model exists. Will have to go back and try.

sequelbox.bsky.social•34 days ago

cool, excited to try it out! glad to see apache license :)

dlsq.bsky.social•34 days ago

Mistral Small 3 architecture is optimised for latency while preserving high quality

dlsq.bsky.social•34 days ago

Mistral Small 3 Base model
https://huggingface.co/mistralai/Mistral-Small-24B-Base-2501

dlsq.bsky.social•34 days ago

Performance of Mistral Small 3 Instruct model
https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501

dlsq.bsky.social•34 days ago

Mistral Small 3 is also available on many partner platforms:
- Ollama: https://ollama.com/library/mistral-small
- Kaggle: https://kaggle.com/models/mistral-ai/mistral-small-24b
- Fireworks: https://fireworks.ai/models/fireworks/mistral-small-24b-instruct-2501
- Together: https://together.ai/blog/mistral-small-3-api-now-available-on-together-ai-a-new-category-leader-in-small-models

And many more soon!

ndrpvt.bsky.social•34 days ago

Congrats!!

Very random question from someone not really too deep on this stuff, but why did you compared it vs Gemma and not Gemini 1.5 Flash?

ndrpvt.bsky.social•34 days ago

(also given that you listed GPT-4o mini)

sorny92.bsky.social•34 days ago

Will the next step be a CoT in the way of Gemini2flash-Thinking.

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply