Here’s what our cloud hosts had to say about this week’s episode: 376: Google Gemini Agents Gemini, Gemini Confused
Justin Brodley
This week on the show we talked about Meta’s new Muse Wonder and Muse Wonder Full models, and they look good, Meta’s own chart has Wonder Full a couple points behind Claude Opus 5.5. They’re also closed, and Wonder Full isn’t even on the API, its Muse Code only. To be fair Meta hasn’t completely walked away from open weights, they shipped Muse Glimmer (30B, Apache 2.0) in August and Muse Spark Open last week. But Spark Open isn’t the model Meta actually serves, it scores 39 on the Artificial Analysis index vs 48 for the version on their API. With Llama you downloaded the real thing, now you get a weaker sibling next to the closed line.
And honestly I’m not sure I want Meta to be America’s open model company anyway, given their track record on privacy and social media safety. Nvidia’s Nemotron 3 Ultra and Reflection’s new Beam are real efforts and i’m glad they exist, but Kimi K3 and GLM 5.3 still score ahead of Meta’s new open model, and Nemotron trailed the previous Kimi. The Chinese models are getting real production use too, Microsoft just built a model on Qwen.
So this is my ask. Will someone please release an open model from a US company that isn’t shady and actually competes with Kimi, GLM, Qwen and DeepSeek? Real weights, a license legal can get through in an afternoon, and the same model you run in production. Please. Come tell me who you think gets there first in the Cloud Pod Slack.
Leave a Reply