GPT-6 Astra built a Mac aquarium wallpaper, battery optimizations pending

AI engineer and educator Chase Lean used OpenAI's GPT-6 Astra to build an interactive aquarium wallpaper for Mac, posting a seven-part thread on X on September 17th showing fish that change direction as the cursor moves. Lean said he is still optimizing the prototype so it would not "eat your battery," that performance depends on the Mac running it, and that a release is coming after further optimization, with no launch date or performance measurements published. OpenAI released GPT-6 Astra on September 3rd with a 1.05 million-token context window, a 128,000-token maximum output, and API pricing of $10 per million input tokens and $50 per million output tokens at standard rates.

GPT-6 Astra built a Mac aquarium wallpaper, battery optimizations pending AI engineer Chase Lean's prototype turns the cursor into fish bait, though performance still depends on the machine running it. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: X https://x.com/chaseleantj/status/2100663203076128908 Why it matters Astra shortened the trip from a one-off idea to working interactive software, while the battery work shows why generated prototypes still need engineering. Chase Lean @chaseleantj https://x.com/chaseleantj , an AI engineer and educator, used OpenAI's GPT-6 Astra to create an interactive aquarium wallpaper for his Mac, producing a desktop scene in which schools of fish change direction as the cursor moves through the water. https://x.com/chaseleantj/status/2100663203076128908 https://x.com/chaseleantj/status/2100663203076128908 Lean posted the experiment in a seven-part thread on X https://x.com/chaseleantj/status/2100663203076128908 on September 17th. The accompanying video shows a virtual aquarium filling the Mac desktop, with fish responding in real time as the pointer moves around the screen. The interaction turns a usually static background into a lightweight simulation rather than another AI-generated image stretched across a display. The prototype remains a work in progress. Lean said he was optimizing it so it would not "eat your battery," and acknowledged that performance depends on the Mac running it. He also said aggressive optimization can improve the result. Those qualifications matter for software that is supposed to remain active throughout the workday: a technically impressive desktop effect loses its appeal quickly if it keeps the graphics processor busy and cuts hours from a laptop charge. Lean indicated that a release is coming after further optimization. His replies also point to additional work on the prototype, though he did not set a launch date or publish performance measurements. The current evidence is the working visual demo and Lean's account of how he produced it. It should be read as a prototype rather than a controlled test of Astra's coding performance. A coding model gets a consumer-sized job OpenAI released GPT-6 Astra on September 3rd https://openai.com/index/gpt-6-astra/ , positioning the model for software engineering, computer use, research and other long-running work. The aquarium is a smaller and more legible demonstration: one person describes a personal software idea, the model produces an implementation, and the remaining work shifts to power consumption, packaging and distribution. That last stretch has long separated a generated demo from software people can keep installed. Cursor tracking and animated fish are familiar programming tasks. Combining them into a polished desktop experience requires decisions about rendering, frame rates, resource use and macOS behavior. Lean's continued optimization work shows that Astra did not remove those engineering constraints. It compressed the path to a working first version. OpenAI's developer documentation https://developers.openai.com/api/docs/models/gpt-6-astra lists a 1.05 million-token context window and a maximum output of 128,000 tokens for Astra. The API costs $10 per million input tokens and $50 per million output tokens at standard rates, with higher charges for prompts exceeding 272,000 input tokens. Lean appears to have worked through ChatGPT, referring another user to GPT-6 Pro in chat, so the public post does not establish what an equivalent API build would cost. The prompt is becoming the product brief The aquarium fits a wider pattern around coding models: bespoke software can now start from requests too narrow to justify hiring a developer or searching for an existing commercial app. A reactive fish wallpaper has limited prospects as a venture-backed company, yet it is exactly the sort of personal utility that demonstrates how demand changes when prototype costs fall. Users can ask for software shaped around a specific device, visual preference or interaction and then decide whether it deserves further work. That expands the addressable set of software ideas far beyond products designed for thousands of paying customers. It also produces a growing pile of programs whose creator still has to handle maintenance, permissions, distribution and hardware differences. Lean's role is central to the result. He has spent years publishing practical experiments with generative AI tools and recently returned to posting after moving to Germany, becoming an AI engineer and pursuing a master's degree in AI. The aquarium continues that approach: use a concrete build to expose both the model's capability and the engineering work left after the first successful run. For OpenAI, experiments like this give Astra a more useful form of marketing than another benchmark table. The model's coding scores matter to engineering teams choosing tools. A cursor chasing fish across a Mac desktop makes the underlying pitch visible to everyone else: custom interactive software is becoming something a user can request before deciding whether it was worth building.