Releases from frontier labs, major companies and research institutes — grouped by model, deduplicated across HuggingFace, OpenRouter and official blogs.
Click a week to filter
DeepSeek🇨🇳Frontier
Multimodal MoE model that natively processes images and text with a 1M-token context window. Based on a 552B backbone, it activates 8B/16B per step, cutting KV-cache overhead to roughly a quarter of its predecessor.
DeepSeek🇨🇳Frontier
Multimodal text-and-image model with substantially improved agent capabilities versus its text-only predecessor; competes in text-only agent tasks but trails leading proprietary models.
DeepSeek🇨🇳Frontier
DeepSeek's heavyweight: 1.6T total, 49B active, 1M context, FP4/FP8 mixed precision. Codeforces 3206 Elo beats GPT-5.4. MIT license, fully open weights. $0.435/$0.87 per MTok on OpenRouter.