ChatGPT can now think with images, handle super long inputs – and it's cheaper than previous versions

0comments
OpenAI logo.
OpenAI is releasing a whole galaxy of new models that are all about getting things done better, faster… and cheaper.

The most recent OpenAI models are called o3 and o4-mini – these are meant to bring powerful upgrades to ChatGPT by letting users solve more complex problems with greater accuracy and flexibility. The two new models can now search the web, read uploaded files, interpret images, write code, and use all tools within ChatGPT.

For the first time, ChatGPT can think with images, not just see them, OpenAI announces. Users can upload whiteboard notes, textbook diagrams, or sketches (even low-quality ones) and the model will understand and analyze them. It can also use tools like Python or browser search when needed, making it far better at tasks like planning, forecasting, or solving multistep problems.

The o3 model is the most advanced and powerful, best suited for in-depth tasks like coding, consulting, and technical problem-solving. It's available to paid ChatGPT users. o4-mini, while smaller and cheaper, still performs impressively and supports higher usage. Neither model is free:


The ChatGPT 4.1 family


Also, OpenAI has released a new family of models:

  • GPT-4.1
  • GPT-4.1 mini
  • GPT-4.1 nano

These models are said to be more capable than previous versions like GPT-4o, especially when it comes to following instructions, writing code, and working with large amounts of text or data. They also come at different price points, making it easier to choose the right model depending on speed, cost, and complexity:


The full GPT-4.1 model is now OpenAI's top-tier option, offering better accuracy in following instructions, solving complex problems, and coding across multiple languages. It can handle extremely long inputs – up to 1 million tokens, which is roughly eight times the size of the entire React codebase (a JavaScript library for building user interfaces). This means developers can now feed in entire projects, massive documents, or multiple sources of information, and the model can still keep track of relevant details without getting confused or lost.

GPT-4.1 is also better at understanding what users want and following custom instructions – like producing a response in a specific format or avoiding certain topics. This makes it more reliable and flexible for building smart applications, such as customer support tools, document analysis assistants, or AI coding partners. It also performs much better than GPT-4o in frontend web development and complex reasoning tasks, its creators claim.

GPT-4.1 mini offers almost the same quality as GPT-4o but with significantly reduced latency and 83% lower cost.

GPT-4.1 nano is the fastest and cheapest model OpenAI has ever released. While smaller, it still delivers surprisingly good performance for tasks like auto-completion, classification, and lightweight assistants. It's ideal for apps that need speed and low cost but don't require advanced reasoning.



All three models are available via API (application programming interface). API access means developers and businesses can connect OpenAI's models – like o3 and o4-mini – directly to their own apps, websites, or software. Instead of using ChatGPT through OpenAI's chat interface, they can build custom experiences where the model works behind the scenes. For example, a company might use o4-mini to power a virtual assistant on their site, answer customer questions, or analyze data in real time.

GPT-4.1 is 26% cheaper than GPT-4o on average, and prompt caching – reusing repeated inputs – now saves even more, which should be great for developers.
Did you enjoy reading this article?
There's more to explore with a FREE members account.
  • Access members-only articles
  • Join community discussions
  • Share your own device reviews
  • Manage your newsletter choices
Register For Free
Loading Comments...
FCC OKs Cingular\'s purchase of AT&T Wireless