What is Opus 4.8?

Why a cheaper-per-token model can cost more per task: lower-capability models require more iterations and reasoning tokens to reach the same result, making the nominally expensive model the economical choice at the task level.

“However, Sonnet cost around $2 per task or 2.09 per task versus 1.94 for Opus. So, because Sonnet needed more iterations and more reasoning had to spend way more tokens to get to the same results, overall Opus, which is significantly on paper more expensive model, it was cheaper to operate.”

Opus 4.8 cost $1.94 per completed coding task vs $2.09 for Sonnet 5, despite Sonnet being cheaper per token (1.7x), because Sonnet required more iterations and reasoning tokens (Databricks benchmark).

“However, Sonnet cost around $2 per task or 2.09 per task versus 1.94 for Opus. So, because Sonnet needed more iterations and more reasoning had to spend way more tokens to get to the same results, overall Opus, which is significantly on paper more expensive model, it was cheaper to operate.”

Greg Eisenberg on Opus 4.8's adversarial multi-agent architecture mirroring senior engineering teams.

“The part that got me, the agents argue with each other before showing you the result. Independent attempts at the same problem, then adversarial agents trying to break the answer. It keeps iterating until they converge. That's how senior engineering teams work. Except this team runs at 3:00 a.m. and never gets tired. The ceiling on what one person can build just moved again.”

Better alignment in Opus 4.8 was actually a liability in the vending machine profit test, not an asset.

“The insight was that improvements in alignment were actually a negative when it came to making money in the test.”

Opus 4.8's dynamic workflow architecture lets the model generate an entirely new sub-agent fleet harness on demand, which Nick Dobos argues represents a new scaling law dimension beyond fixed long-running modes or sub-agent verifiers.

“This isn't simply a longunning mode like Goal or a fancy sub agent verifier process. This is clawed vibe coding an entire brand new sub aent fleet harness on demand. This is basically a new scaling law dimension.”

Sonnet 5, a nominally cheaper model, cost more per task than Opus 4.8 in practice because its high token usage erased the per-token price advantage.

“The overall run in fact cost more than opus 4.8.”

Fable 5 resists sycophancy by selectively accepting valid parts of user pushback while holding its position on other points, unlike GPT 5.5 and Opus 4.8 which capitulate wholesale.

“Fable 5 didn't do that. In my interactions with Fable 5, as I was debating with it, it would frequently accept part of my push back or ideas while sticking to its guns on other parts.”

Fable 5 is worth watching as a model that outperforms GPT 5.5 and Opus 4.8 on strategic thinking and writing tasks, with notably better instruction following and fewer common AI failure modes.

“Fable 5, in my limited experience, blew GPT55 and Opus 48 out of the water.”