Major

Opus 5.5 is insane

Opus 5.5 cut the cost of a Major agent session by 79% compared to Opus 5.0. Here's where the savings come from and why cheaper models are good for a hosting platform.

Jason BaoUpdated
Bar chart: Opus 5.5 vs. Opus 5.0 per Major agent session, with 41% fewer agent steps, 40% fewer output tokens, 71% less cached context re-read, and 79% lower cost.

Background

There are two ways to build apps and workflows on Major - you can use Major through a plugin or you can interact directly with the Major agent.

Today, a large percentage of our customers still develop with the Major agent because it has been harness-optimized to interact with our stack (+ the experience is just really nice 😊).

Opus 5.5

Opus 5.5 has reduced costs per session 79% compared to Opus 5.0.

Bar chart of Opus 5.5 against Opus 5 per user message: 41% fewer agent steps, 40% fewer output tokens, 71% less cached context re-read, and 79% lower cost.

That reduction comes from two areas:

  • Fewer tokens
  • Cheaper tokens
Waterfall chart of cost per message from Opus 5 to Opus 5.5: 2.5x less from fewer tokens, then 1.9x less from cheaper tokens.

What this has led to is more people building on Major, spending a fraction of the cost.

Weekly sessions and spend from Sep 7 to Sep 28: sessions grew after Opus 5.5 rolled out on Sep 21 while weekly spend dropped.

Our stance

We are thrilled about this. Major is not in the business of reselling tokens. We are a hosting platform - we grow when customers build and deploy useful applications. We make money when you build things that matter. Cheaper models mean more builders, more apps in production and a better Major.