SWE-2 Max just cooked up this pirate ship at sunset, entirely as an .html file, right inside Devin CLI.
🏴☠️🔥 cognition has been cooking HARD with their new models. I became a fan of SWE-1.7 this month, and now we already have SWE-2. And honestly? In this pirate-ship faceoff,
Voxel pagoda test with Deepseek 4.1, generally at the level of an Opus 5
SITUATION EXPLAINED: DeepSeek V4.1 Flash beats its own Pro model from 3 weeks ago.
552B mixture of experts, 8B active for input and 16B for output. MIT licensed, 1M context • Beats V4 Pro on every benchmark except GPQA Diamond and Humanity's Last Exam • $0.15 input, $0.60
I just compared DeepSeek-v4.1-Flash against Opus 5
Gave same prompt to both models to build ocean sandbox using three.js DeepSeek-v4.1-Flash was almost 90% cheaper and 50% faster than Opus 5 💀 maybe the benchmarks are not broken.. do you think DeepSeek is better?
You have reached the end of the archive
All of Claude Opus 5