Meta Platforms has officially launched its most powerful AI model to date, Muse Spark 1.3, which focuses on extending agent workflows and enhancing coding capabilities. Alexandr Wang, Meta's Chief AI Officer, stated that the model has surpassed OpenAI's GPT-5.6Sol in core programming abilities and is on par with Anthropic's Claude Fable5.1.
In terms of pricing strategy, Muse Spark 1.3 maintains the standard of its predecessor, Muse Spark 1.2, offering paid access to developers through the Meta Model API. The specific cost standards are: $1.25 (approximately 8.4 RMB) per 1 million token input; $0.15 (approximately 1 RMB) per 1 million token cache hit input; and $4.25 (approximately 28.6 RMB) per 1 million token output. According to Alexandr Wang, the API platform market is currently experiencing strong growth, with some top developers using up to trillions of tokens per week. In the future, this new model will be gradually integrated into Instagram, Facebook, and the Meta AI assistant.
Compared to the previous generation, Muse Spark 1.3 has achieved significant optimization in code generation quality. It greatly reduces unnecessary code iterations, producing more concise code with a clearer overall coding style. In practical tool call performance, the number of calls has decreased by about 20%, and the number of tokens required to complete the same task has also been reduced by 25%.
To better handle complex scenarios, Muse Spark 1.3 has been deeply optimized for long-term agent tasks. It can collaborate on multiple workflows within a single long thread, proactively correct plan gaps, and track learning outcomes. In terms of user experience, when faced with ambiguous instructions, the model will actively ask for clarification; when encountering obstacles, it will request user assistance; and before performing critical operations, it will seek human confirmation.
