New models and developer products announced at DevDay

OpenAI used its first DevDay to reset the terms of its developer platform, and the headline is GPT-4 Turbo. The new model reads up to 128,000 tokens of context, which works out to roughly 300 pages of text in a single prompt — a huge jump from the 8K and 32K windows developers were stuck juggling before. OpenAI also cut prices, making input tokens cheaper and output tokens cheaper still, a move clearly aimed at anyone who found GPT-4 too expensive to run at scale.

Vision comes along for the ride too. GPT-4 Turbo with Vision lets the model take images as input alongside text, so a developer can now build a single application that reads a screenshot, a photo, or a scanned document and reasons about it in the same conversation as everything else. Pair that with the new DALL·E 3 API, and OpenAI is quietly building out an image pipeline that goes both directions: understand pictures, generate pictures, all through the same account and billing.

The more structurally interesting release is the Assistants API. Instead of developers hand-rolling memory, retrieval, and tool-calling logic every time they want to build something more useful than a plain chatbot, OpenAI is offering a framework that handles persistent threads, file retrieval, and code execution out of the box. It's a tacit admission that most people building on GPT-4 were solving the same plumbing problems over and over, and OpenAI decided to just ship the plumbing.

Taken together, these aren't flashy consumer features — they're infrastructure moves. Longer context, lower prices, native vision, and a scaffolding layer for agents are the kind of changes that shift what an indie developer or small team can realistically ship without stitching together five different tools. That's the real story of DevDay: OpenAI consolidating its lead not by making a smarter model, necessarily, but by making the whole stack cheaper and less annoying to build on.