Visit TestSprite: https://www.testsprite.com/?via=aicodingki-dw
TestSprite CLI (Github Repo): https://github.com/TestSprite/testsprite-cli
In this video, I’ll be testing Meta’s new Muse Spark 1.2 model and its brand-new terminal coding agent, Muse Code. We’ll look at the new coding upgrades, pricing, Muse Code features, and then run Muse Spark 1.2 through KingBench 3 to see how well it performs against models like Fable 5, Qwen 3.8 Max, Opus 4.8, Opus 5, Kimi K3, and more.
—
Key Takeaways:
🚀 Meta has released Muse Spark 1.2, a coding-focused upgrade to Muse Spark 1.1.
💻 Muse Code is Meta’s new terminal coding agent, designed to work closely with Muse Spark 1.2.
🧠 The model has been trained heavily on long-horizon coding tasks, full repositories, and end-to-end projects.
⚙️ Muse Code includes async background agents, restart-safe runs, event logs, and useful slash commands.
💸 The pricing includes a cheaper data-sharing tier, plus $20 of free starting credit.
📊 Muse Spark 1.2 scored 61 out of 80 on KingBench 3, landing in fifth place overall.
🎨 The model performed especially well on frontend, SVG, 3D, and visual coding tasks.
🛠️ Backend logic and complex simulations still need improvement compared to the top frontier models.
👍 Overall, Muse Spark 1.2 is a strong release and a big improvement over Muse Spark 1.1.
source
