Meta 发布 Muse Spark 1.1 模型
阅读原文· the-decoder.comMeta 推出 Muse Spark 1.1 模型。据 Artificial Analysis 数据，该模型在 Intelligence Index 上得分 51，与 GLM-5.2、GPT-5.4 和 GPT-5.6 Luna 持平；Coding Index 得分 71.3，领先 GLM-5.2（68.8），略低于 GPT-5.6 Luna（71.4）。每任务成本约 0.26 美元，低于 GLM-5.2（0.37 美元）和 GPT-5.4（0.89 美元），输出 token 使用量仅 9400 万。模型幻觉率从 73% 降至 38%，上下文窗口扩展至 100 万 token。发布初期仅通过 Meta 自有 API 提供。
Meta's new Muse Spark 1.1 model edges ahead of GLM 5.2 in coding while coming in at a lower price point. According to Artificial Analysis, it scores 51 on the Intelligence Index, tying with GLM 5.2, GPT-5.4, and GPT-5.6 Luna. In just three months, the model gained eight points, mostly in coding and agent-based knowledge work. On the Coding Index, it scores 71.3, ahead of GLM 5.2 (68.8) and barely behind GPT-5.6 Luna (71.4). The top spots belong to GPT-5.6 Sol (77.4) and Terra (76.7), followed by Claude Fable 5 (76.5). As always, benchmark scores don't always match real-world performance.
Muse Spark 1.1 costs an estimated $0.26 per task, compared to $0.37 for GLM-5.2 and $0.89 for GPT-5.4, while using only 94 million output tokens (GLM-5.2 uses 141 million). The hallucination rate dropped from 73 to 38 percent: the model now more often declines to answer rather than giving wrong ones. Meta also quadrupled the context window to one million tokens. At launch, Muse Spark 1.1 is available only through Meta's own API.