88Z AI, 3 September — Meta has released its new large language model, Muse Spark 1.3. Its coding capabilities surpass GPT-5.6 Sol and match Claude Fable 5.1. Optimized for long-cycle Agent tasks, the model can proactively correct plans and ask clarifying questions. API pricing remains at previous levels, with tool calls reduced by approximately 20% and token consumption lowered by 25%.
At a glance
- Coding capabilities surpass GPT-5.6 Sol and match Claude Fable 5.1
- Tool calls reduced by 20% and token consumption down 25% with unchanged pricing
The update
Meta has officially launched Muse Spark 1.3, its most powerful AI model to date. In terms of coding performance, the model excels, surpassing GPT-5.6 Sol and matching the capabilities of Claude Fable 5.1. Muse Spark 1.3 is deeply optimized for long-cycle Agent tasks, enabling it to proactively correct execution plans, ask clarifying questions, and seek user confirmation, thereby improving task accuracy. Regarding cost and efficiency, the new model maintains the same API pricing strategy as its predecessor. Through technical optimizations, tool calls have been reduced by approximately 20%, and token consumption has decreased by 25%. This allows developers to access more efficient and intelligent model services without increasing their budget, significantly lowering long-term operational costs.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- IT之家Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.