Faster, but the homework is not finished. The same sentence prompts for glaciers, aurora and titles.
DeepSeek and GLM ran to 150 and 180 tok/s on four Spark rings, but they didn't do everything. Qwen3.8-27B only has 61 on Mac Studio, but it can produce all three. Warm reminder: A single comparison of the same set of tips by the author. Source: @CryptoBandit100👇
Bitter lesson on the robot side?
Foundation models and dedicated robot models might be getting one from frontier models like Claude Fable 5.2, GPT-6 Astra, and Gemini Argon...(soon, Deepseek) Fable 5.2 hit 64.9% on RoboDojo’s bimanual sim benchmark -- more than double its 5.1
Lebrel AI season 1 full update ~ free AI every day + badges + $lbrl points 😱💰
LebrelAI closed the waitlist, opened beta and moved earning into First Run Season 1 - missions now give Season XP, badges and daily free sessions. ⇨ Status: Season 1 Live ⇨ Free Window: 3-7 PM
One account. Leading AI models, all in one place.
GPT, Claude, Gemini, DeepSeek, Kimi, Qwen, GLM & Grok—no separate sign-ups or repeated setup. Switch models within the same task in Moxus Desktop. Less setup. More creating.
You have reached the end of the archive
All of deepseek