This is all you need to know about Science's Last Exam (tl;dr)
113 executable research tasks across 9 fields (34 institutions). Best overall score is 37.5/100 Opus 5. 36.15/100 GPT-6.1 Sol. All 8 frontier models scored 0 on true end-to-end research.
"Say Good morning to Claude at 7 AM."
We measured it in Claude Code: Sonnet 5.5 burns 30K tokens, Opus 5.5 16K. 1.9x. That tip from ParamSiddh (652K views): send Claude a "Good morning" at 7 so your 5-hour window resets two hours into your workday. Nobody says what the greeting
This is f*cking insane. Andrej Karpathy
"The biggest mistake in AI right now - people are forcing agents to work instead of mastering the harness engineering for model first We made that mistake in 2016 at OpenAI - It cost us 5 years " Opus 5.5 → Prompts → Agents → Graphs
A making-of video of a music video using Suno MCP x Opus 5.5 to create a song and MV using only the environmental sounds of a coin laundry.
Of course, there is sound, so please watch with the sound on! The snare is the sound of a T-shirt being ripped off, the kick is the sound of a door closing, and the melody is the beep of a button. All accompaniment is played using only sounds generated by AI videos.
You have reached the end of the archive
All of Claude Opus 5