Alexandr Wang announces Muse Spark 1.3 Max with coding and agentic gains

Alexandr Wang posted on X that his company has released Muse Spark 1.3 Max to the public. The new version sits atop a tiered release that began with 1.3 High and 1.3 XHigh. Wang describes Max as a significant step up in coding and agentic performance — the model's ability to plan and execute multi-step tasks autonomously. He recommends testing it even for users who have already evaluated the earlier tiers.

The announcement does not specify what distinguishes the three tiers technically. It is unclear whether Max is larger, trained longer, or fine-tuned differently. The company offers only a general claim of "significantly stronger performance."

No benchmarks accompany the release. Wang cites no scores on HumanEval, SWE-bench, Arc, or other standard evaluations, and provides no quantitative comparison against High or XHigh. The claims remain unverified manufacturer statements.

Wang says the model is available now and encourages hands-on testing. The post does not explain how to access it — whether via API, chat interface, open weights, or under what commercial terms.

Also absent: a knowledge cutoff date, known limitations, context window size, pricing policy, and any mention of internal or external safety evaluation. For developers considering production deployment, these omissions carry as much weight as the advertised improvements.