Claude Opus 5.5: Cheaper AI model builds games in one go

What's new

On September 22, 2026, Anthropic launched a new version of its AI model, Claude Opus 5.5. The model is the first in a new 5.5 series, and according to Anthropic it is the strongest model the company has ever tested. It is already available to all users on the Claude platform as well as via Amazon, Google and Microsoft's cloud services (AWS, Google Cloud and Azure). In a few weeks, smaller and cheaper versions, Claude Sonnet 5.5 and Haiku 5.5, are also coming.

What's clever

The most striking thing is that users have gotten the model to build entire, playable 3D games from a single sentence, without fixing a single line of code afterward. In several published examples, Opus 5.5 itself wrote 3D graphics, sound and animations in code, without using ready-made images or sound files from outside, and put the game directly online so it could be played right away. In other words, the model can both plan and make finished videos and small games from scratch, a task that previously required an entire team of developers and designers.

Cheaper or better?

Yes to both. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens (tokens are the pieces of text the AI computes in). That is 20 percent cheaper than its predecessor Opus 5, and for repeated requests (cache) the price has dropped a full 60 percent, from $0.50 to $0.20 per million tokens. At the same time, the model responds more than 30 percent faster. It is Anthropic's own estimate that Opus 5.5 overall costs around 40 percent less to run than Opus 5 on typical tasks, because it solves them with fewer tokens and fewer rounds. So it is not a fixed discount: how much you save depends, among other things, on how long the answers are, how much is cached, and how much the model thinks. On several programming tests, it performs better than both Opus 5 and the competitor Fable 5.1; for example, it scores 66.4 percent on a test called Terminal-Bench, versus 55.8 percent for Fable 5.1. For regular Pro, Max, Team and Enterprise subscribers, there are also higher usage limits, and you can now save a "reset timer" option to use when it suits you best.

What it's good at

Opus 5.5 is especially strong at what Anthropic calls agentic coding, meaning AI that plans and carries out an entire task on its own, such as building a program from start to finish, instead of just answering a question. It is also good at knowledge work such as reports and analyses, as well as understanding images and diagrams. For companies, this means you can get prototypes, videos or small programs built faster and more cheaply than before, without putting an entire team on the task.