Meta Platforms Inc. is preparing to release a new version of its Muse Spark artificial intelligence model, with a focus on advanced programming capabilities. Alexandr Wang, the company’s chief AI officer, said on X that the update will roll out “soon.”
Benchmarks suggest big gains over previous models
Reporters stated that the new algorithm is competitive with GPT-5.5 across several “closely followed” AI benchmarks, though the report’s sources didn’t name which ones. Most frontier model announcements include results from SWE-Bench Pro, a standard test for coding ability.
Related: Jamf debuts Beacon threat hunting service for Mac
The original Muse Spark scored 52.5% on that benchmark.
GPT-5.5, the OpenAI model that Meta’s new algorithm reportedly matches, reached 58.6%. The OpenAI model also outperformed the earlier model on another popular programming benchmark called Terminal-Bench 2.0.
Wang said on X that Meta’s upcoming model is significantly more adept at code generation than its predecessor and also better at powering AI agents. The current model includes a “contemplating mode” that uses AI agents to improve prompt responses. During internal testing, the company had the model complete a test called HLE with and without the feature. It scored 8% higher when contemplating mode was enabled.
Related: Arcade gets 60 million dollar boost
An X user asked Wang when the company would launch a model that can match the programming capabilities of Anthropic’s Claude Opus 4.8. The executive responded that it will happen “pretty soon.” Claude Opus 4.8 scored 69.2% on SWE-Bench Pro, or 10.2% higher than OpenAI’s flagship model, though it fell behind that model on Terminal-Bench 2.0.
Higher performance comes with higher infrastructure costs
The improved output quality of Meta’s new model reportedly comes at the expense of increased infrastructure use. According to the article, the algorithm uses an “order of magnitude” more computing capacity than the earlier model. That trade-off suggests the company is prioritizing raw performance over efficiency in this release.
Related: Global tensions fuel demand for risk intelligence
The company’s push to improve software development capabilities may signal plans to make these models accessible to external developers. On Wednesday, reporters noted that the Facebook parent is considering launching an AI infrastructure service. Executives are reportedly debating whether it should offer raw computing capacity or hosted AI models. Taking on existing tools such as Claude Code will require the company to build more than just code-optimized models. Claude Code offers integrations with popular developer tools, a desktop app, and customization features. Users can also configure it to repeat a task at specific time intervals.
Code generation isn’t the only use case it could target with its planned successor to the current model. Anthropic offers Claude Code alongside Claude Cowork, a productivity tool geared toward nontechnical professionals. That offering includes several vertical-specific feature bundles for industries such as healthcare and financial services. The social media giant hasn’t announced any similar product plans, but the competitive market suggests a broader platform play may be coming.
