这些是读者照片,与图钉带来源的媒体分开显示。
第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
Meta Newsroom 的文章标注为“Originally published on April 8, 2026 at 9:00AM PT”,开头为“Today we are announcing Muse Spark”;Artificial Analysis[2][5] 和 OfficeChai[6] 在同一天发布了各自的分析。
Follow-up that uses Muse Spark 1.0 as the baseline: 43 on the later index version, Coding Index 59, GDPval-AA v2 1,144, 262k context, against 51 for Muse Spark 1.1 three months later.[1]
Context piece: Meta's first model in a year and the first product of a nine-month-old Superintelligence Labs, a natively multimodal reasoning model with instant, thinking and contemplating modes, inputs up to 262,000 tokens, undisclosed size, free in meta.ai and the app, API preview for select partners, and developer concern after Meta left open weights.[1]
Context: MSL was set up last June after the $14.3bn Scale AI deal and the hiring of Alexandr Wang, cut 600 jobs in October, and Meta worked with more than 1,000 physicians on health training data; Meta says it performs on par with GPT-5.4, Claude Opus 4.6 and Gemini 3.1 Pro, with no numbers given.[1]
Independent launch analysis: 52 on the Intelligence Index behind Gemini 3.1 Pro, GPT-5.4 and Claude Opus 4.6 and ahead of Claude Sonnet 4.6, GLM-5.1, MiniMax-M2.7 and Grok 4.20; Meta's first frontier model not released as open weights; 58M output tokens to run the index; Humanity's Last Exam 39.9%, MMMU-Pro 80.5%; weaker on GDPval-AA (1,427) and Terminal-Bench Hard.[1]
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: Muse Spark Released
| 基准 | Muse Spark | 对比 |
|---|---|---|
| CharXiv Reasoning (Thinking) | 86.4 | Gemini 3.1 Pro 80.2; GPT 5.4 82.8 |
| HealthBench Hard | 42.8 | GPT 5.4 40.1; Gemini 3.1 Pro 20.6 |
| DeepSearchQA | 74.8 | Gemini 3.1 Pro 69.7; Grok 4.2 62.8 |
| Humanity's Last Exam,不使用工具(Contemplating) | 50.2 | Gemini 3.1 Deep Think 48.4; GPT 5.4 Pro 43.9 |
| ARC AGI 2 (Thinking) | 42.5 | Gemini 3.1 Pro 76.5; GPT 5.4 76.1 |
| Terminal-Bench 2.0 | 59.0 | GPT 5.4 75.1; Gemini 3.1 Pro 68.5 |
| GDPval-AA Elo | 1,444 | GPT 5.4 1,672; Opus 4.6 1,606 |