The codex-1 model is state of the art on SWE-Bench Verified, we published the numbers on the blog. More importantly, we optimized it to generate code that people actually want to merge, not just code that scores well on benchmarks!
MACHINE LEARNING
-
AI Tutoring Programs Boost STEM Higher-Order Thinking
By
–
– Programs lasting 4-8 weeks were the most effective
– Learning performance rose the most in problem-based learning
– AI acting as a tutor vs. tool drove more higher-order thinking gains
– AI boosted higher-order thinking for STEM content/courses most over other types -
OpenAI Codex Agent Shows Net-Positive Coding Gains, Future Bright
By
–
I've tried OpenAI's Codex, the autonomous SWE agent integrated into ChatGPT. -> While certainly not producing one-shot perfect PRs, it's already a net-positive. And the gain will only improve with better coder models! I think that in the next era of coding, most will be done
-
Technology Failures: Incremental Limits vs Fundamental System Constraints
By
–
Most of the time, it doesn't matter that some technology can't do something yet, because you can tell it's only a matter of incremental refinement. But sometimes, you should pay close attention, because the failure tells you something important about the nature of the system.
-
Codex 1: Parallel Tasks on OpenAI Compute
By
–
Codex 1 will be a new model that can run many tasks in parallel on OpenAI's Compute
-
AWS-Powered AI Enhances Ferrari F1 Performance at Fiorano
By
–
Still buzzing from an unforgettable day on Ferrari’s Fiorano circuit. Grateful for the experience, the people, and the tech. AWS-powered data helped us improve every lap—human performance, augmented by AI. This is innovation in motion. #Ferrari #AWS #AI #F1Technology #GenAI
-
Anthropic develops Claude Code IDE extensions
By
–
Anthropic seems to be working on Claude Code extensions for VSCode and JetBrains IDEs. Competition for vibe coders h/t @flikteoh
-
AI Inference Costs Drop Enable Advanced Agentic Workflows Viable
By
–
“None of this would have been possible even a year ago. But with the cost of AI inference dropping, context windows expanding, reasoning models handling much more complex tasks, and better understanding of designing agentic workflows, this all becomes viable.”
-
Engineering Behind Dancing Robots: Code, AI and Hardware
By
–
Que un robot baile parece simple, pero detrás de cada paso hay miles de líneas de código, sensores que corrigen en tiempo real, motores que simulan equilibrio humano y una coordinación milimétrica entre hardware y software.
— Juan Merodio (@juanmerodio) 16 mai 2025
Esto no es solo un robot bailando.
Es ingeniería, IA y… pic.twitter.com/wUrlOnHYk8Que un robot baile parece simple, pero detrás de cada paso hay miles de líneas de código, sensores que corrigen en tiempo real, motores que simulan equilibrio humano y una coordinación milimétrica entre hardware y software. Esto no es solo un robot bailando.
Es ingeniería, IA y
