My FREE AI OS Course: https://www.skool.com/ai-automation-society/about?el=opus-4.8-release&hcategory=youtube-videos&utm_campaign=free-group Full courses + unlimited support: https://www.skool.com/ai-automation-society-plus/about?el=opus-4.8-release&hcategory=youtube-videos&utm_campaign=ais-plus Apply for my YT podcast: https://podcast.nateherk.com/apply Work with me: https://uppitai.com/ My Tools💻 FREE MONTH voice to text: https://get.glaido.com/nate Code NATEHERK for 10% off VPS (annual plan): https://www.hostinger.com/vps/claude-code-hosting Release blog: https://www.anthropic.com/news/claude-opus-4-8 Prompting docs: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices Opus 4.8 just dropped and the benchmarks are honestly nuts, but the numbers only tell part of the story. In this video I break down my key takeaways after reading through Anthropic's own documentation, and how I'm thinking about slotting Opus 4.8 into my actual workflows. If you want to actually get the most out of this model instead of just running it the same way you ran 4.7, start here. Sponsorship Inquiries: 📧 [email protected] Connect with me: https://www.linkedin.com/in/nateherkelman/ https://x.com/nateherk https://www.instagram.com/nateherk/ TIMESTAMPS 0:00 Intro 0:35 What's New in 4.8 1:07 Effort Levels and Workflows 2:05 Benchmarks Reality Check 2:54 The Honesty Upgrade 4:38 4.7 Pain Points 6:52 Key Takeaways 10:33 Community Reactions 12:12 Final Thoughts
ADVERTISEMENT
Where in the school community do I find the token tracker link?
How does one know when brainstorming and GetShitDone are only dragging us back compared to the most recent models?
It’s a sly trap. How do I know which particular task actually needs medium, high or extra high? Isn’t that very subjective? I may think it’s an easy task but I get inferior output because Claude thinks it’s not easy? It creates more confusion and frustration - having one more option to consider, and where one never knows for sure if that’s the right option.
I never even migrated to 4.7 -- my CC still reports 4.6 as the highest... so what do I need to do? tried /model claude-opus-4-8 and it just enables opus 4
Thanks for diving into this. Yes, we needed this update, but funny thing, I actually like 4.7's snakiness. I found it refreshing, actually. There are even times when it detects that I'm procrastinating on something related to what I'm chatting with it about, and it begins using language that helps snap me out of it. Yes, it's a bit much at times, but I left OpenAI a long time ago because of its tendency towards sycophancy.
It's worth noting that while it costs the same as 4.7, a lot of the time it thinks for longer and uses more tokens as a result
u got glyph drops too?
in 5.0 xhigh + workflows will be the lowest setting.
had a full session with opus 4.8 at high setting for an header issue. the shit did not resolve it. ~danny
I'll wait for the fireship video
i feel that the honesty is something all ai should do, ai should stop making claims if not verified and stop jumping to conclusions. This is how I felt when i used sonnet 4.6
You only "played with it for half an hour and wanted to get this video out quick"? Man. I need to rethink watching your videos.
What happend to Claude Mythos?
This will completely revolutionize data centers around the country. Build them right on top of those Southern homesteads at this point tbh.
Please do the same review on the Sonnet 4.6 😎
At this point it seems they only do a little tunning, mostly to fix errors that they introduced in 4.7 and just launch the patched version as a new one.
Is Karpathy's influence noticeable?
Nate I continue to love your content. Any chance you can do a video on your workflows and tools used to get your content out at pace after the news drops.
What about them starting to drive this in the path of efficiency because it costs way too much now. I get that it helps a lot but the cost is going out of hand and for the cost difference DeepSeek or Kimi will just do the same job...
Love you, Nate. I got your video before Anthropic's email by 1 hour. LOL