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 AI Research的文章'Introducing Muse Spark 1.1'日期为2026年7月9日,开头为'Today, we're excited to introduce Muse Spark 1.1';TechCrunch[5]当天报道了此次发布,而OpenRouter[2]在7月16日的上架记录是该路由器添加该模型的时间。
Router listing 'Meta[1]: Muse Spark 1.1': created 2026-07-16, 1,048,576-token context, text, image, video and file input, $1.25 per 1 million input tokens, $4.25 per 1 million output tokens and $0.15 per 1 million cached input.
Independent comparison: coding index 71.3 against 68.8 for GLM-5.2 and 71.4 for GPT-5.6 Luna, 94M output tokens against 141M for GLM-5.2, $0.26 per task against $0.37, and a hallucination rate down from 73% to 38% because the model declines to answer more often.[1]
Independent benchmarks: 51 on the Intelligence Index (up 8 from Muse Spark 1.0's 43), Coding Index 59 to 71, GDPval-AA v2 1,144 to 1,376 Elo, hallucination rate 73% to 38%, about $0.26 per index task, 94M output tokens, 1M context up from 262k, $1.25 and $4.25 per 1 million tokens, about 114 tokens per second.[1]
Same-day report by Lucas Ropek: Meta[1] released the model publicly on Thursday 9 July for agentic coding, priced at $1.25 and $4.25 per 1 million tokens in line with Claude Haiku 4.5 and GPT-5.6 Luna, with Zuckerberg calling it 'a strong agentic and coding model at a very low price' while Meta trails Anthropic and OpenAI.
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 1.1 Released