Muse Spark 1.1 编程超越 GLM-5.2 且成本降低 0.11 美元
meta 的 muse spark 1.1 在编程基准测试中以 71.3 分击败 glm-5.2，单次任务成本仅为 0.26 美元，标志着高性价比编码代理的兴起。
meta upgraded its muse spark model to version 1.1, which now scores 71.3 on artificial analysis's coding index, surpassing glm-5.2's 68.8 and nearing gpt-5.6 luna's 71.4. priced at $0.26 per task, it undercuts competitors like glm-5.2 ($0.37) and gpt-5.4 ($0.89) while using significantly fewer output tokens. the update also slashed the hallucination rate from 73% to 38% and expanded the context window to one million tokens. these improvements position muse spark as a highly efficient option for enterprise coding workflows, prioritizing reliability and lower inference costs.