Botterill says Opus 5 feels more commonsense than GPT-5.6 Sol on instruction following
JasonBotterill · x · 2026-07-26
Replying to @dejavucoder, Jason Botterill says Anthropic does unusually well at strong base models and that its non-thinking Opus 5 feels highly capable, even though it has not been broadly benchmarked.
He later says that after using Opus 5 more, he finds GPT-5.6 Sol overly strict in instruction following, while Opus 5 better captures what the user actually wants and is more forgiving when prompts are messy.
More from Fun
- Quackchestra turns prompts into MIDI files with a chat interface and playable output — BLCNYY · 2026-07-26
- Claude keeps defending Opus 5 instead of writing a critical review — sethlazar · 2026-07-26
- Codex gets roped into booking a camping trip for a fake disabled dad — yacineMTB · 2026-07-26
- A researcher meme perfectly captures the fury of missing citations — docmilanfar · 2026-07-26
- Backrooms-style AI short film explains the hand behind the avatar — charis_ai · 2026-07-26
- A short Doom-and-Reed meme post lands as an in-group joke — WellsLucasSanto · 2026-07-26