Tovie AI’s Illustration Re-imagined by ChatGPT
OpenAI has been busy. Over the past month, it rolled out a series of major updates, including two new models, o3 and o4 mini, that promise smarter reasoning and better logic handling. Alongside them comes the GPT 4.1 family, including mini and nano versions, plus an enhanced image generation tool in GPT 4o that works like a text-based Photoshop on steroids. Here’s what these updates mean and who they are designed for.
GPT 4.1 gets more memory and better coding skills
The GPT 4.1 models can now process up to one million tokens in context, about 750,000 words. Output length has also doubled to 32,000 tokens or around 24,000 words, though OpenAI recommends caution when using full capacity as accuracy may drop when memory is overloaded. These are general-purpose models, but their strongest suit is code. GPT 4.1 shows improved performance across development tasks and can now handle long documents and multi-step queries more efficiently. The model comes in three versions:
- GPT 4.1 (flagship): Best suited for complex tasks with large or layered inputs. Ideal for developers and professionals working with detailed data or processes.
- GPT 4.1 mini: A cheaper and faster option that keeps strong coding accuracy but is less effective with large-scale inputs.
- GPT 4.1 nano: Lightweight and fast. Useful for autocomplete, data extraction and simple classification tasks.
Independent analysts report that GPT 4.1 outperforms the earlier GPT 4o by a wide margin, 21 per cent better at coding, 7 per cent better with long context handling, and at least 10 per cent more efficient at multitasking. None of the GPT 4.1 models are available in the ChatGPT interface. They are accessible only via API with pricing ranging from 10 cents to 2 dollars per million input tokens and 40 cents to 8 dollars per million output tokens, depending on the model.
o3 and o4 mini: built for smarter reasoning
Two other models, o3 and o4 mini, bring a different focus. These models are designed to think before they speak with stronger logic and better handling of complex questions that require reasoning over multiple steps. Both models can use external tools like web search or plug into custom APIs, choosing when to access them. For example, they can analyse a company’s sales database, retrieve live inventory data and create a logistics plan, all in a single conversation.
They also work well with images. Whether interpreting graphs or analysing photos, o3 and o4 mini can reason about visual data, for example, identifying a star formation and determining the hemisphere in which the photo was taken.
- o3 has the highest reasoning capacity and supports up to one million tokens. It is ideal for maths problems, academic discussions or data heavy analysis. However it is not always accurate and may reinforce its own mistakes. OpenAI reports that it makes errors in about one third of its responses, twice the rate of earlier reasoning models.
- o4 mini is a faster and more affordable alternative with a shorter memory window, about one-fifth that of o3. It is suitable for simpler tasks like summarising product reviews or drafting marketing text.
At present o3 is only available to ChatGPT Plus users and through the API, priced at 10 dollars per million input tokens and 40 dollars per million output. o4 mini is more affordable at 1.10 and 4.40 respectively, and is partially accessible to free users via the new Reason mode in ChatGPT.
At the end of March, OpenAI quietly upgraded GPT 4o’s image generator and the change is already making waves. Originally reliant on DALL-E for visual output, GPT 4o can now generate and edit images directly within ChatGPT, even on the free tier.
Users can request images, change styles, adjust details and add text, not just as captions but within the visuals themselves. The tool works iteratively, so users can refine images step by step without starting over. Social media was quick to react. Creators began sharing anime-style screenshots, Studio Ghibli characters and stylised film frames. Even official bodies took part: the Royal Mail and the White House both used generated images, though the latter sparked debate by publishing a synthetic image of a migrant detention centre.
This upgrade is proving useful across creative industries. Designers are using it for concept mockups, marketers for visual A/B testing, and entrepreneurs for rapid prototyping and product imagery. The generator’s output is now much more accurate. It produces fewer distortions, more realistic hands and faces, and better spatial layouts. It can also edit real photos, for example changing signage in a street scene or design an interior layout based on a single reference image.
- GPT 4.1 adds longer memory and better performance for developers
- o3 and o4 mini offer improved reasoning and tool integration
- GPT 4o’s visual tools empower creators with rapid image generation and editing
Transform your business with Generative AI