Somewhat of a surprising result with this GLM 5.3 Flash (running in OpenCode) spreadsheet cleanup test.
The task was to clean 1,001 rows of typical "manually entered" spreadsheet stuff. One prompt, one attempt (because trust is the metric that matters for this stuff). Ran the
Improved the King Kai planet mission today for my DBZ runner game!
Built w/ my game dev harness usecrayon, threejs and Opus 5!
Claude Opus 5 vs GPT-5.6 Terra for computer use.
Build one self-contained HTML page titled SHADOW SHOW and save it as shadow-show.html. Show a giant red-velvet puppet… Claude Opus 5 won against GPT-5.6 Terra. Go try computer-use agents for free at
The two best coding harnesses are in a dead tie.
The harness is the whole story. this video nails it. agent = model + harness. change only the harness and the result moves more than the model ever would. same benchmark, same sonnet, different harness — different agent. senior
You have reached the end of the archive
All of Claude Opus 5