Today, @AnthropicAI dropped Opus 4.7, their most powerful model yet. For the past week, we’ve been stress-testing it at Zapier.
This isn't the "Mythos" model you've heard about - but it's a meaningful step up.
Early findings from our benchmarks:
1. Modest step up from Opus 4.6. ~20% improvement in accuracy, for about the same cost
2. When you use max reasoning effort, it's the highest scoring model we've tested to date
3. It's more efficient than other models - if you're batching multiple tool calls in agentic workflows, this is where it shines. Same with multistep workflows
4. The one downside: it can be overly cautious. If you tell it a task is really important, it might refuse to act rather than risk being wrong
Overall: 20% more accurate than Opus 4.6, same cost. Just comes with added caution.
Try Opus 4.7 now in @Zapier.
@lennysan Zapier's hosting an AMA with our Head of RevOps next week—she'll break down our AI Fluency Framework and how we use it in practice: https://t.co/Avj425c6zU
It's official! We're launching OpenAI on Zapier 🎉
Thank you, @yoheinakajima, for building something amazing that we can now share with everyone!
P.S. Y'all are already building SUPER cool stuff with AI + Zapier.
Here are a few of our favorites ⬇️
Here's my response to a recent article about how work from home kills careers. Thankfully, the reality is much more promising—esp. at the best companies.
https://t.co/Ug6JBxKZ7x