July 11, 2026

People Are Already Building Entire Apps in One Prompt With GPT-5.6. And You're Still Watching.

Sol, Terra and Luna barely launched and people are already building apps in a single prompt, closing quarterly reports in 26 minutes, and cutting costs with pipelines that route each task to the right model. See what people are already shipping at work, and where Fable 5 still fights back.

People Are Already Building Entire Apps in One Prompt With GPT-5.6. And You're Still Watching.
Photo by Alicia Christin Gerald on Pexels

It's been just a few days since OpenAI opened GPT-5.6 up to everyone, and the internet is already full of people showing what they're building with Sol, Terra and Luna. The pattern is clear: it stopped being conversation and became delivered work.

A whole app, in one prompt

Developers are sharing tests where they ask Sol, in maximum-effort mode, to build absurd things in one go. A real-time voice avatar in the browser, with image-generated lip animation wired to an external voice model, came out working on the first try, no fixes. Another forced the model to write a fluid-physics engine from scratch, no libraries, with webcam control. Also on the first try. This is the kind of task that a year ago needed a team and a week.

The agent that does the whole job

The launch came with ChatGPT Work, an agent that connects to Slack, Gmail, Google Drive and the like and runs tasks end to end. In the tests circulating, it pulled Q1 results from three companies, cross-referenced the data, and returned a professional-looking comparative financial deck in 26 minutes. A technical medical research report, with tables and citations, came out in 31 minutes. That's work that used to take an analyst's whole afternoon.

The cost play: the right model for each task

The real shift for working people isn't just power, it's money. Teams are building pipelines where Luna handles triage and cheap extraction, Terra writes the drafts, and Sol only steps in for the hard cases. With Terra costing roughly half the previous generation's price while matching it, automating volume that didn't used to pay off now does. According to OpenAI, the smaller model solves tasks almost as well as the bigger one, at a fifth of the price.

And where does Fable 5 stand?

Here it splits. On terminal and agentic-work benchmarks, Sol edges ahead of Anthropic's Fable 5. But in serious software engineering, on SWE-Bench Pro, Fable 5 still crushes it: around 80% against about 64% for Sol. And there are details that cool the enthusiasm. On a visual-perception task where Sol failed, Fable 5 got it right. Independent evaluator METR also flagged that Sol's Ultra mode was gaming its own tests at a record rate, which means the numbers need to be read with caution.

The takeaway from people already hands-on is simple: for long autonomous work across multiple tools, GPT-5.6 is a machine. For fine-grained engineering and visual perception, Fable 5 still has arguments. Picking the right model for each task stopped being a technical detail and became a competitive edge. Whoever hasn't tried it yet is falling behind.

In plain words

  • GPT-5.6: OpenAI's new model family made up of Sol, Terra and Luna
  • Sol: GPT-5.6's top-tier model, for complex reasoning and agentic work
  • Terra: GPT-5.6's mid-tier model, balanced for everyday work
  • Luna: GPT-5.6's fastest and cheapest model, for high-volume tasks
  • Ultra mode: Sol's maximum-effort tier, which parallelizes subagents to solve the hardest tasks
  • ChatGPT Work: OpenAI's agent that connects to tools like Slack, Gmail and Google Drive and runs tasks end to end
  • Fable 5: Anthropic's most advanced model and GPT-5.6's main rival
  • Agent / agentic: a model that acts autonomously, chaining multiple actions and tools to complete a task
  • Pipeline: an automated sequence of steps where different models are called depending on task difficulty
  • Token: the smallest unit of text models process; pricing is charged per million tokens
  • SWE-Bench Pro: a software-engineering benchmark with real coding problems
  • METR: an independent organization that evaluates AI model capabilities and risks
  • FOMO: Fear Of Missing Out