Opus 4.8 vs Opus 4.7
and GPT-5.5

Opus 4.8 is a convincing improvement from Opus 4.6 and Opus 4.7, and is now officially the default model in Shortcut. We will be deprecating support of Opus 4.6 and 4.7.

We had our issues with Opus 4.7 in both benchmarking and taste-based internal sessions, and never made the switch from Opus 4.6 to Opus 4.7 for our default model.

It felt like Opus 4.7 was trained with a heavy emphasis on dynamic effort, allowing it to be more forceful in choosing how long and hard to think about tasks in relation to their complexity. Unfortunately, its judgment for spreadsheet complexity was off. Opus 4.7 would shortchange easier tasks, getting outscored by Opus 4.6 on most realistic Excel work. Opus 4.7 was better on extremely difficult tasks, including our v25 benchmark below, which is composed of the hardest industry spreadsheet work in the world, but only when run on high reasoning. It would also take far longer at high reasoning. Its place on the Pareto curve did not feel like a strong user experience.

During this time, OpenAI released GPT-5.5 and firmly re-established itself as a contender, and every bit as good as Anthropic in the Excel for AI space. GPT-5.5 is clearly the smartest public model in the world right now, but that advantage shows up primarily on the most extreme spreadsheet tasks. We have not made it the default for two main reasons. First, it still lacks the Opus-level polish and taste required for a great Excel model to feel intuitive, clean, and well-designed. Second, Opus 4.8 remains state of the art on our everyday benchmark, v23, which better represents the day-to-day spreadsheet work done by even the world's top investment firms.

Opus 4.8 corrects nearly all of Opus 4.7's shortcomings. It has much stronger decision-making when it comes to how hard and long to think on a spreadsheet task. We recommend Opus 4.8 at Medium effort, which scores as high as Opus 4.7 did at High effort in about half the time. Fast Mode on Opus 4.8 is also 3x cheaper than Fast Mode was for Opus 4.6/4.7, meaning it is finally economically viable for your most time-sensitive work.