Benchmark Gains and Performance Metrics

Gavin details benchmark improvements: coding, agentic tasks, and long-horizon autonomy where Opus 4.7 outperforms previous releases.

Play episode from 02:28

chevron_right

Transcript

chevron_right

Transcript

Episode notes

Anthropic dropped Claude Opus 4.7 with better vision, better coding and… better everything. And, along with OpenAI's new Codex, AI is accelerating ever faster.

This week on AI For Humans, Anthropic released Claude Opus 4.7, a major step up from Opus 4.6 with better visual reasoning, improved software coding and even makes presentations for cavemen.

Benchmarks put Opus 4.7 between 4.6 and the unreleased Mythos preview, and the new default xhigh reasoning level means more token burn but more reliability on hard problems. The same day, OpenAI updated Codex with better computer use, an integrated browser, and a bunch of new tools.

Then Jensen Huang's epic Dwarkesh Patel interview broke the internet, with Jensen explaining why NVIDIA keeps selling AI chips to China and dropping the instantly iconic "you're not talking to someone who woke up a loser" line.

Plus, Reese Witherspoon is now in on AI, Doug Liman's Killing Satoshi got made for $80M using AI tools (would have cost $300M without them) and we got our first look at AI Val Kilmer.

OPUS 4.7 HAS LANDED. CODEX GOT UPGRADED. IT'S ALL HAPPENING

Come to our Discord: https://discord.gg/muD2TYgC8f

Join our Patreon: https://www.patreon.com/AIForHumansShow

AI For Humans Newsletter: https://aiforhumans.beehiiv.com/

Join our TikTok @aiforhumansshow

To book us for speaking, please visit our website: https://www.aiforhumans.show/

// Show Links //

Claude Opus 4.7 Official Blog Post

https://www.anthropic.com/news/claude-opus-4-7

Claude Opus 4.7 System Card

https://cdn.sanity.io/files/4zrzovbb/website/037f06850df7fbe871e206dad004c3db5fd50340.pdf

Opus 4.7 Is Better at Presentations