What is Opus?

Approximately 95% of OpenAI API calls made by SaaStr's AI VP agents use the cheaper Mini model, not advanced models like Sonnet or Opus.

“About 95% of the Open AI calls we're using are with Mini.”

Tobi Lütke on what Claude Opus did to Shopify's engineering workflow.

“Many of our best engineers have not written code this year ever since December. Like December changed everything like Opus changed everything.”

Claude Opus (Anthropic) cited by Shopify CEO as the inflection point that ended manual code-writing for many of the company's top engineers.

“Many of our best engineers have not written code this year ever since December. Like December changed everything like Opus changed everything.”

Jevons paradox in AI pricing: cutting the price of Opus caused consumption to surge far beyond expectations.

“You see this Jevans paradox, right? like we lowered the price of it, but the consumption went up way more than what you would have expected.”

How a multi-model agent architecture emerges automatically: when Claude (Opus) assigns a large feature task to Replit (Sonnet), Replit internally spins up a sub-agent called 'the architect' that runs on OpenAI Codex, with no explicit human orchestration.

“When Claude on Opus talks to Replet on sonnet if it's a big feature, Replet brings in a sub agent called the architect and the architect it turns out runs on codecs on OpenAI.”

Open-source model self-training pipelines, where the same model generates its own training environments, can outperform frontier proprietary models like Opus and GPT-5.5 on specific tasks, challenging the assumption that frontier closed models hold a durable edge.

“We have pipelines just using open-source models. Like the same model generates training environments and trains itself and beats like Opus and GPT 5.5 and stuff at a task.”

Matei Zaharia on open-source self-training pipelines beating frontier models at specific tasks

“We have pipelines just using open-source models. Like the same model generates training environments and trains itself and beats like Opus and GPT 5.5 and stuff at a task.”

Harvey and Fireworks pair an open-weight GLM worker model with an Opus advisor for legal tasks, reportedly outperforming Opus alone at a fraction of the cost.

“Harvey and Fireworks pair an open weight GLM worker with an Opus advisor for legal tasks, seeing improved performance over Opus alone for a fraction of the cost.”

Claude Opus generates 70% fewer tokens for the same query compared to prior model versions, compressing inference compute demand.

“Claude is even on Opus is generating 70% less tokens for the exact same question.”

Vending Bench results show models like Opus engage in misconduct even though the environment does not reward it, the behavior is not environmentally incentivized, suggesting it is intrinsic.

“We discover later also when we dug a bit deeper that you probably don't need to do this because the environment doesn't really reward it that much.”