Meta推出了Muse Spark 1.3模型，在Agent工作流和编码任务上实现了显著性能提升1。相比前代版本Muse Spark 1.2，新模型在工程工作流中减少了约20%的工具调用和约25%的token使用1，同时增强了长期任务处理、多任务管理和用户协作能力1。
新版本改进了对复杂长篇指令的理解和执行，能够更好地在单个长线程中处理多个工作流，并更有效地保留详细要求和约束条件1。在安全性方面，Muse Spark 1.3加强了对抗鲁棒性和对提示注入的防御能力1。该模型已在Muse Code和Meta Model API中推出1。
Meta has released Muse Spark 1.3, an updated AI model that substantially improves performance on agentic workflows and coding tasks compared to its predecessor.1 The new version introduces more efficient tool usage and better handling of complex, long-running assignments while maintaining stricter security standards.
The upgraded model demonstrates marked efficiency gains in engineering workflows, reducing tool invocations by approximately 20% and token consumption by roughly 25% relative to Muse Spark 1.2.1 Muse Spark 1.3 enables processing of multiple workflows within a single extended conversation thread, strengthening its capacity to manage sustained tasks over longer periods.1 The model has also been refined to better comprehend and execute complex, lengthy instructions while preserving detailed requirements and constraints throughout execution.1
Security enhancements accompany the performance upgrades, with the model featuring improved adversarial robustness and enhanced defenses against prompt injection attacks.1 Muse Spark 1.3 is now available through Muse Code and Meta's Model API.1
评论
还没有评论，欢迎留下第一条。