MiniMax 发布开源权重模型 M3，支持 1M 上下文与原生多模态，具备前沿编码与 Agent 能力
多源视角
M3 would never 🙂↔️<br><br>As a matter of fact, the weights are now open, too.<br><br>https://huggingface.co/MiniMaxAI/MiniMax-M3<hr style="border:0;border-top:1px solid #80808030…
查看这条来源M3 is running together 🤝 with @togethercompute, and with faster-than-ever inference<hr style="border:0;border-top:1px solid #80808030;margin:12px 0;"><div class="rsshub-quote">Tog…
查看这条来源the kernels are doing the lord's work today, day-0 on @vllm_project, verified on nvidia and amd. <br><br>go read the writeup 👇<hr style="border:0;border-top:1px solid #80808030;ma…
查看这条来源day-0 in @vllm_project and it comes with:<br><br>dedicated MSA prefill/decode kernels, 1M-context serving with prefix caching + chunked prefill, BF16 + MXFP8 on both Hopper and Bla…
查看这条来源M3 now on @FactoryAI droid<hr style="border:0;border-top:1px solid #80808030;margin:12px 0;"><div class="rsshub-quote">Factory: MiniMax M3 has arrived in Droid.<br><br><img width="…
查看这条来源With only ~428B params, and ~23B activated params<br><br>M3 still handles frontier coding + long-horizon agents + native multimodal (text, image, video) at 1M-token context<br><br>…
查看这条来源appreciate it @SambaNovaAI 🤝 looking forward to M3 on RDUs<hr style="border:0;border-top:1px solid #80808030;margin:12px 0;"><div class="rsshub-quote">SambaNova: Congrats to our p…
查看这条来源RT GMI Cloud<br>MiniMax M3 is live on GMI<br><br>M3 is the first open-weight model combining frontier coding & agent capabilities, 1M-token context, and native multimodal under…
查看这条来源means a lot coming from @NVIDIAAI <br><br>free GPU-accelerated M3 endpoint are live now<br><br>go try it 👇<hr style="border:0;border-top:1px solid #80808030;margin:12px 0;"><div c…
查看这条来源AI 摘要
MiniMax 发布开放权重模型 MiniMax M3。该模型总参数量约 428B、激活参数量约 23B，支持 1M-token 上下文窗口与原生多模态理解，并具备前沿编码与 Agent 能力。模型权重已上架 Hugging Face，同时发布了 MiniMax Sparse Attention 论文。开发者可直接下载部署，在超长上下文、多模态及 Agent 场景中进行应用开发与评测。 核心观点： 1. MiniMax M3 总参数量约 428B，激活参数量约 23B，以开放权重形式发布。 2. 模型支持 1M-token 上下文窗口，具备原生多模态理解及前沿编码与 Agent 能力。 3. 权重已上架 Hugging Face，并配套发布 MiniMax Sparse Attention 技术论文。
推荐理由高信息密度，值得细读
原文
无法获取正文，可打开下方原始链接查看。
讨论
暂无评论。