Sonoma (Grok 4 Mini) & Qwen-3 Max: 2 New Models are here but are they GOOD? скачать в хорошем качестве

Sonoma (Grok 4 Mini) & Qwen-3 Max: 2 New Models are here but are they GOOD?

Visit NinjaChat: https://ninjachat.ai/ In this video, I’ll be testing the new stealth models Sonoma Dusk and Sonoma Sky, plus Qwen Max, with hands-on demos, benchmarks, and honest thoughts on whether you should use them. We’ll cover context windows, tool calling, KiloCode setup, pricing guesses, and better alternatives like GLM, DeepSeek, and Kimi K2. -- Key Takeaways: 🆕 Two new “Sonoma” stealth models are here: Dusk and Sky, likely related to X-AI’s Grok family. 🌤️ Sky performs better than Dusk; both sometimes do reasoning and claim a 2M-token context window. 📊 Leaderboard: Sky ranks 10th and Dusk 15th—okay, but behind GLM and other top models. 🧪 Hands-on results are mixed: floor plans don’t make sense, Panda SVG is mediocre, and chess autoplay makes illegal moves. ⚙️ Coding/tool-calling are decent and very fast; KiloCode integration is easy—if you try one, pick Sky. 🤖 Agent test (movie tracker) asked for a real TMDB key and hit reliability/rate-limit issues; output felt lackluster. 🚫 Qwen Max is API-only and not open-sourced; it ranked 12th, struggles with tool calling, and over-thinks/crashes on math. ✅ Better options today: GLM, DeepSeek, and Kimi K2; Long Cat is underrated; nothing here is game-changing—waiting on Gemini 3. -- Timestamps: 00:00 - Introduction 00:12 - About Sonoma (Grok 4 Mini) 01:33 - NinjaChat (Sponsor) 02:23 - Leaderboard results & Usage of Sonoma 04:55 - Qwen Max overview and API-only status 06:33 - Final thoughts + waiting for Gemini 3 07:48 - Outro