Grok 4.5 helped improve already top-10 kernel competition solutions
tokenbender · x · 2026-07-26
Two users independently tried Grok 4.5 on separate kernel competitions and both managed to improve already highly optimized top 5–10 solutions.
The poster says that makes the model feel promising for future versions and deserves more recognition.
More from Models
- Claude builds a Brawl Stars-style game from scratch with AI-generated art, gameplay, and music — minchoi · 2026-07-27
- Google Reportedly Launches Gemini 3.6 Flash Models with Improved Coding — dl_weekly · 2026-07-27
- Ant Group launches Ling-3.0-flash with 124B parameters and a free API until Aug. 3 — Loose_Bank1709 · 2026-07-27
- Reddit users say Gemini 3.5 Pro's new checkpoint is "deadass insane" — Last_Conclusion_8984 · 2026-07-26
- Reddit debate says benchmark wins still do not predict daily LLM usefulness — Alternative-Car8221 · 2026-07-26
- Reddit user says ChatGPT daily limit has dropped from 100 messages to 5 — Visual-Gas6540 · 2026-07-26