Meta promises open-weights Muse Spark model soon, unveils upgraded 1.3 version

Mark Zuckerberg posted on X that Meta plans to release an open-weights version of its Muse Spark large language model in the near future. Until that rollout, the company has made Muse Spark 1.3 live on its API service and the Muse Code CLI, highlighting improvements such as fewer unnecessary token uses, clearer responses to ambiguous queries, and more frequent prompts for human assistance. Meta claims these changes lower the number of interaction turns and overall cost, while benchmark results from Artificial Analysis indicate a four-point intelligence boost, matching top competitors like GPT 5.6 Sol and Claude Opus 5.

The new version also offers a contributor pricing tier with rates as low as $0.002 per million cached input tokens, $0.10 per million input tokens, and $0.20 per million output tokens. Since the original Muse Spark launch in April, Meta has issued several refinements to demonstrate progress and justify its AI-related capital spending. The upcoming open-weights release is positioned as a move to increase transparency and foster broader adoption.