Apr 22, 2026
312 Research Blog
Image models for teaching: what the benchmarks miss
GPT Image 2 launched yesterday. The benchmarks are loud. But teaching puts a different stress test on image models than anything the launch reels optimize for — here's what actually matters.
AO
Adilet Osmonkulov
Apr 16, 2026
312 Research Blog
I built an AI coworker on Tuesday. I'm rebuilding him on Thursday.
A short story about setting up an AI coworker named Sheldon with OpenClaw, watching his identity drift from his files, and deciding to rebuild on Claude Managed Agents two days later.
MS
Marsel Shaibekov