Alexandr Wang's team is releasing the weights for coding and long-horizon tasks, with local operation requiring two M5 Ultra Mac Studios or substantial GPU memory.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
· Updated
Primary source: [X](https://x.com/alexandr_wang/status/2110778736434184611)
Why it matters #
Meta is putting Wang's frontier-model work into developers' hands as downloadable weights, but the hardware demands and API-specific license threshold distinguish access from unrestricted deployment.
Meta released the weights for Muse Spark Open on October 15th, giving developers access to an 816-billion-parameter multimodal model designed for coding and long-running agent tasks. Alexandr Wang, who leads Meta's AI effort, described it as the most capable US open-weight model to date; that ranking is Meta's claim, not an independently established result.…
source & further reading
runtimewire.com — original article
Liquid AI releases d1-ultra for multimodal decisions at $0.042 per million input tokens
Anthropic gives six METR evaluators internal access and publication rights
FTC investigates Anthropic and OpenAI over AI safety incidents