Muse Spark 1.3
Points and comments are a snapshot, not live.
Meta releases Muse Spark 1.3, trained for agentic workflows and competitive coding.
Muse Spark 1.3 is trained for long-horizon, agentic workflows and competitive coding performance, with higher first-attempt accuracy and reliable tool calling. It tracks context and prior results, handles messy inputs, and asks for clarification when needed. The model offers native multimodal perception of video, images, and documents, with visual reasoning in a real execution environment. Pricing: contributor model $0.10/M input tokens, $0.20/M output; standard $1.25/M input, $4.25/M output. Context window is 1M tokens.
What commenters are saying
Commenters focused on the pelican-on-a-bicycle SVG benchmark, with several showing 1.3 outputs. One user found 1.3 produces better bicycle frames, wings, and pelican hats than 1.2, but noted it costs about 3x the tokens. A common observation: pelicans nearly always face right, likely because bicycle product shots and drivetrain placement favor that side. Some speculated the benchmark may be gamed by training on pelican SVGs, while others argued general improvements show it is not saturated. A few criticized the model for giving pelicans arms instead of using wings as arms.