New OpenAI models — major improvements on coding, instruction following, and long context

This is from Open AI

Hey there,

I wanted to check in and give you an update on what’s new at OpenAI. Over the past month, we’ve launched three new models offering significant improvements over their predecessors. In particular, GPT-4.1 represents a major upgrade over GPT-4o, especially for real-world software engineering tasks—achieving a 55% score on SWE-bench Verified compared to GPT-4o’s 33%.

A quick overview of what’s new:

We released our latest flagship GPT family. GPT-4.1, 4.1 mini, and 4.1 nano are our first models trained for developer use-cases, featuring substantial improvements in coding, instruction-following, and function-calling. GPT-4.1 models have larger context windows—supporting up to 1 million tokens—and better utilize this expanded context with improved comprehension. They’re significantly more skilled at frontend coding, suggesting cleaner diffs, structured responses, reliable tool use, and more. Additionally, these models are 26% cheaper than GPT-4o on median queries, and our first nano model is our fastest and most cost-effective model ever. Get started with GPT-4.1.

Our newest reasoning models, o3 and o4-mini, are now available. o3 achieves leading performance in coding, math, science, and vision, while o4-mini provides faster, cost-efficient reasoning. In the Responses API, both models support reasoning summaries—streaming the model’s thought process as you await the final response—and enable smarter tool usage by preserving the model’s prior reasoning between calls. Get started with o3 and o4-mini.

I’ve seen many developers adopt these models recently and recognize their coding capabilities, instruction-following, speed, and price, so I hope’ll you get a chance to try them out too! And as always, please let us know what you think.

Best,

Edwin Arbus

Developer Experience

0 comments