Meta Platforms Inc. is preparing to roll out a powerful new version of its Muse Spark artificial intelligence model imminently.
Chief AI Officer Alexandr Wang announced the upgrade is coming ‘soon,’ following reports the algorithm matches the performance of OpenAI’s GPT-5.5 on several major AI benchmarks. Internally, the company highlights a substantial leap in software development capabilities.
On the SWE-Bench Pro coding test, the original Muse Spark scored 52.5%. GPT-5.5 achieved 58.6%, while Anthropic’s Claude Opus 4.8 leads the field at 69.2%. Although Meta’s new model aims to catch up, it still trails the latest Claude iteration on certain specialized tasks. Wang responded to user inquiries by stating a model competitive with Claude Opus 4.8’s programming skills will arrive ‘pretty soon.’
The performance gains require a massive increase in computing resources, reportedly using an order of magnitude more infrastructure capacity than its predecessor. This move aligns with reports that the Facebook parent is exploring a new cloud business to sell excess AI compute power.
The enhanced model is not just a coding tool. It also improves the efficacy of its ‘contemplating mode,’ which uses internal AI agents to boost the quality of prompt responses by as much as 8% in internal testing. This positions Meta to compete directly in the market for professional productivity and developer-centric AI agents.