Kimi K2 got a massive upgrade, possibly the best open source coding model now?
I think Kimi K2 is incredibly good, overtime we'll see how it compares with Qwen 3 Coder. But I showcase what I've learned, and discuss pricing, demos, and prompt caching. Links: π§βπ»My Recommended AI Engineer course is Scrimba: https://scrimba.com/the-ai-engineer-path-c02v?via=GosuCoder My Links π ππ» Subscribe: https://www.youtube.com/@GosuCoder ππ» Twitter/X: https://x.com/GosuCoder ππ» LinkedIn: https://www.linkedin.com/in/adamwilliamlarson/ ππ» Discord: https://discord.gg/YGS4AJ2MxA My computer specs GPU: RTX 5090 (sometimes a AMD 7900xtx) CPU: 7800x3d RAM: DDR5 6000Mhz Media/Sponsorship Inquiries β gosucoderyt@gmail.com
Video Chapters
- 0:00 Kimi K2's Latest Evolution: A Deep Dive into Enhanced Context and UI
- 1:22 Unleashing the Power: Kimi K2's Context Window Soars to 262K Tokens!
- 8:50 The Speed Frontier: Groq's Blazing Fast Performance vs. OpenRouter's Limits
- 11:24 Smarter Spending: How Prompt Caching Slashes Your API Costs
- 15:32 Code Quality Deep Dive: Kimi K2's Refined Output
- 17:09 Project Spotlight: Kimi K2 Redesigns a Website for Just 74 Cents!
- 20:25 The Physics Challenge: When Kimi K2 Stumbles on a Pool Game
- 22:38 Beyond Expectations: Kimi K2 Builds a Fully Functional Live Chat App
- 23:47 The Real Cost: What a Month of Active Groq Coding Could Set You Back
- 26:01 Kimi K2's Final Verdict: Strengths, Weaknesses, and Future Outlook
Original Output
0:00 - Kimi K2's Latest Evolution: A Deep Dive into Enhanced Context and UI 1:22 - Unleashing the Power: Kimi K2's Context Window Soars to 262K Tokens! 8:50 - The Speed Frontier: Groq's Blazing Fast Performance vs. OpenRouter's Limits 11:24 - Smarter Spending: How Prompt Caching Slashes Your API Costs 15:32 - Code Quality Deep Dive: Kimi K2's Refined Output 17:09 - Project Spotlight: Kimi K2 Redesigns a Website for Just 74 Cents! 20:25 - The Physics Challenge: When Kimi K2 Stumbles on a Pool Game 22:38 - Beyond Expectations: Kimi K2 Builds a Fully Functional Live Chat App 23:47 - The Real Cost: What a Month of Active Groq Coding Could Set You Back 26:01 - Kimi K2's Final Verdict: Strengths, Weaknesses, and Future Outlook Timestamps by StampBot π€
Unprocessed Timestamp Content
0:00 - Kimi K2-0905 update focuses on improving context window and frontend abilities. 0:20 - Original Kimi K2 struggled with small context window and slow speed. 1:22 - Big update: Kimi K2's context window doubled to a perfect 262k tokens. 2:15 - Vibe Check: Kimi K2 often creates new files during refactoring, surprisingly useful. 3:07 - UI Redesign Example: Initial Kimi K2 design of Knowledge Base had too much purple. 3:59 - Redesign Take Two: Kimi K2's second attempt at the Knowledge Base, double sidebar. 5:10 - Personal Portfolio Design: Old Kimi K2 generated a smooth, dark-themed portfolio. 5:54 - New Portfolio Attempt: Latest Kimi K2 generated a jumpier, light-themed portfolio. 7:45 - Design Comparison: Old portfolio seems better, less "jumpiness" in its design. 8:50 - Speed & Providers: Groq is fast, but OpenRouter's Groq implementation suffers rate limits. 9:55 - Turbo Version: Moonshot AI Turbo offers good speed and prompt caching. 10:37 - Groq in OpenCode: Direct Groq usage in OpenCode encountered unexpected context errors. 11:24 - Cost Analysis: Prompt caching drastically reduces API costs for repetitive tasks. 12:47 - Context Lengths: Groq direct often reports lower context use, potentially saving money. 15:32 - Quality Eval: Kimi K2's coding quality is marginally better than previous versions. 16:12 - John Doe's Portfolio: Another Kimi K2 design with some non-functional input boxes. 17:09 - Gosu Evals Redesign: Kimi K2 successfully redesigned the website for just 74 cents. 17:27 - Mobile Marvel: The redesigned Gosu Evals site is now impressively mobile-responsive. 19:38 - Arena Shooter: A surprisingly fun game generated, but can't defeat enemies! 20:25 - Pool Game Blues: Kimi K2 struggles with accurate physics for a web-based pool game. 21:05 - Virtual Pet: A basic pet simulator with cute sleeping animations and health stats. 21:39 - Drone Simulator: Features procedural generation and customizable drone controls for navigation. 22:38 - Real-time Chat: A working live chat app was generated, including historical messages. 23:47 - Groq's Monthly Bill: Estimating over $300 for a month of active Groq coding. 26:01 - Kimi K2 Insights: Good context window, but performance and pricing still pose challenges. 27:00 - Latency Logs: Moonshot AI logs show variable 'Time to First Token' due to queueing. 29:29 - Pool Game Perfection: Python version of pool game has decent physics, just needs better holes! Timestamps by StampBot π€