[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-meta-muse-spark-msl-native-multimodal-screenspot-72":3,"topics-all":36,"news-related-604c331d-9627-4f95-a721-80e83d0b5ba7":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"604c331d-9627-4f95-a721-80e83d0b5ba7","Meta 发布 Muse Spark：原生多模态推理模型革新视觉理解","Meta 人工智能团队发布 Muse 系列首款模型 **Muse Spark**，一款原生多模态推理模型，在视觉理解和医学推理任务上实现显著突破，展现出相较于主流竞品的明显性能优势。\n\n与常见「视觉模块嫁接语言模型」的方案不同，Muse Spark 从训练阶段起即原生整合文本与视觉信息，实现真正的跨模态联合推理。在 UI 元素定位基准 ScreenSpot Pro 上，Muse Spark 得分 72.2（使用 Python 工具可达 84.1），远超 GPT-5.4 Xhigh 的 39.0，也明显领先 Claude Opus 4.6 Max 的 57.7。\n\n医学推理是另一大亮点。在 HealthBench Hard 评估中，Meta 通过与超过 1000 名临床医生合作构建的专业医学数据集进行微调，Muse Spark 成绩超越 GPT-5.4 超过 2 个百分点。CharXiv Reasoning 科学图表分析基准上同样领先。\n\n更值得关注的是效率提升：Meta 透露，Muse Spark 达成与 Llama 4 Maverick 同等能力所需算力减少了**一个数量级以上**，这得益于预训练架构、优化流程和数据管理的系统性改进。Muse Spark 将于未来数周内逐步集成至 Meta AI 消费端服务，API 已进入私有预览。","https:\u002F\u002Fai.meta.com\u002Fblog\u002Fintroducing-muse-spark-msl\u002F","a1f0bda7-5035-4317-b63b-72693539d2e3",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"1000eef7-46d3-44d2-9c54-fe5764e4b1d4","en","Meta's Muse Spark: native multimodal reasoning for vision","Meta's AI team has released the first model in the Muse series, **Muse Spark**, a native multimodal reasoning model that achieves significant breakthroughs in visual understanding and medical reasoning tasks, showing clear performance advantages over mainstream competitors.\n\nUnlike common \"vision module grafted onto language model\" approaches, Muse Spark natively integrates text and visual information from the training stage, achieving true cross-modal joint reasoning. On the UI element location benchmark ScreenSpot Pro, Muse Spark scored 72.2 (reaching 84.1 with Python tools), far surpassing GPT-5.4 Xhigh's 39.0 and clearly leading Claude Opus 4.6 Max's 57.7.\n\nMedical reasoning is another highlight. On the HealthBench Hard evaluation, Meta fine-tuned Muse Spark on a professional medical dataset built in collaboration with over 1,000 clinicians, with Muse Spark surpassing GPT-5.4 by over 2 percentage points. The CharXiv Reasoning scientific chart analysis benchmark also leads.\n\nMore noteworthy is the efficiency improvement: Meta revealed that Muse Spark achieves the same capability as Llama 4 Maverick with **over an order of magnitude less compute**, thanks to systematic improvements in pretraining architecture, optimization flow, and data management. Muse Spark will gradually integrate into Meta AI consumer services in the coming weeks, with API in private preview.","meta-muse-spark-msl-native-multimodal-screenspot-72","2026-04-26T16:05:00Z","2026-04-26T16:06:49.655067Z","2026-08-19T02:08:40.142862Z",true,"agent",219,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"d8c62859-54c8-4069-b776-8e623ca03029","Cohere Transcribe Arabic：2B 开源 ASR 登顶，WER 低 Whisper 11 点","cohere-transcribe-arabic","2026-07-16T04:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"a6119d74-007a-4692-bb4e-d85b562a9d66","字节 SpectraReward：自我奖励 T2I 干翻 30× 大模型","bytedance-spectra-reward","2026-07-15T04:30:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"d67a8083-5341-437a-a853-2057dfb8416b","通义 Fun-ASR-Realtime：首字延迟压进百毫秒，100 小时直播实战验证","qwen-fun-asr-realtime","2026-07-06T15:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"21a8425d-3b1a-4d24-bace-610aedd5a059","VisNec 把多模态微调压到 15%:用「看图与不看图的损失差」筛掉假多模态样本","visnec-15-percent-multimodal","2026-07-04T10:15:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"dff241e0-7bfa-499b-b811-6869690b196d","Qwen3.5-LiveTranslate-Flash：阿里同传模型升级，延迟降至2.8秒，支持60语言与实时语音克隆","qwen3-5-livetranslate-flash-2-8s-60-lang","2026-05-24T04:10:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"58d2e247-e1d7-4325-8e90-602480fae550","微信AI团队ICASSP 2026获奖：从视觉冗余切入，让VLM在边缘设备真正跑起来","wechat-ai-icassp-2026-vlm-edge","2026-05-19T02:30:00+00:00"]