Asanshay Gupta
173 posts





We are sharing an early preview of our ongoing SWE-1.6 training run. It significantly improves upon SWE-1.5 while being post-trained on the same pre-trained model - and it runs equally as fast at 950 tok/s. On SWE-Bench Pro it exceeds top open-source models. The preview model still exhibits some undesirable behaviors like overthinking and excessive self-verification, which we aim to improve. We are rolling out early access to a small subset of users in Windsurf.




I have been on this setting since 10am and running things in the same session all morning (literally zero breaks except when I got up for food). I've been coding/prompting 12 hours+ straight. I expected context bloat, context rot after multiple auto-compact. There has been zero drift. What have you guys made?!



pencil autocomplete #3 realtime model: FLUX.2 [klein] by @bfl_ml via @fal







Introducing PayPilot - AI-native payroll & HR for modern teams. We built the HR platform we wished existed: - AI assistant that actually answers employee questions - Run payroll in minutes, not hours - Beautiful UX (yes, HR software can be delightful) paypilot-one.vercel.app








