Mistral News 09/11/2024 AI Rating: Medium

Mistral AI Releases Its First Multimodal Model, "Pixtral 12B"

#Mistral#Mistral AI#AI History

On September 11, 2024, Mistral AI released its first multimodal model, “Pixtral 12B.” Built on the text model “Nemo 12B,” it stands out for its ability to handle any number of images of any size (via URL or base64 encoding), and was released under the Apache 2.0 license, freely available for anyone to use and modify.

Details

  • Size: 12 billion parameters, with a model size of about 24GB
  • Multimodal capabilities: Designed to handle image understanding tasks similar to those of the Claude family and GPT-4o, such as generating captions for images or counting objects in a photo
  • Input format: Accepts images via URL or base64 encoding, with no limit on the number or size of images
  • License and distribution: Released under the Apache 2.0 license via a torrent link on GitHub and on Hugging Face, with no restrictions on downloading, fine-tuning, or commercial use
  • Platform integration: Sophia Yang, head of developer relations, said on X that it would soon also be available to try on Le Chat and La Plateforme

What happened next

Pixtral 12B later evolved into “Pixtral Large” that November, marking the point at which multimodal capability became firmly integrated into Mistral’s model lineup.

Source https://mistral.ai/news/pixtral-12b/

Related Articles

More Mistral articles →
Mistral News 07/24/2024 Read AI Rating: Medium

Mistral AI Announces "Mistral Large 2" — Claiming Parity With GPT-4o and Llama 3 405B

Mistral AI announced its second-generation flagship, "Mistral Large 2." At 123 billion parameters, it claims performance on par with GPT-4o, Claude 3 Opus, and Llama 3 405B.
Mistral News 02/26/2024 Read AI Rating: High

Mistral AI partners with Microsoft, announces "Mistral Large" and chat app "Le Chat"

Mistral AI has announced its flagship model "Mistral Large" and conversational app "Le Chat." The company also revealed a partnership with Microsoft, which is investing in Mistral AI and becoming its primary cloud partner via Azure.
Mistral News 12/11/2023 Read AI Rating: High

Mistral AI releases "Mixtral 8x7B" — an open MoE model rivaling GPT-3.5

Mistral AI has released "Mixtral 8x7B," a mixture-of-experts model that outperforms Llama 2 70B while running six times faster, delivering GPT-3.5-level performance with open weights.
Mistral News 09/27/2023 Read AI Rating: High

Newcomer Mistral AI unveils "Mistral 7B" — enters the scene by outperforming much larger models

Newly founded Mistral AI unveiled the fully open "Mistral 7B," drawing attention for outperforming both models in its own size class and the far larger Llama 1 34B.