When new Opus or GPT models were released, they initially performed better on the TerminalBench b..., Sonic AI
“When new Opus or GPT models were released, they initially performed better on the TerminalBench benchmark within Factory's Droid than in their native harnesses like Claude Code or Codex.”