This chart from METR shows LLM task success rates for a longer task duration. The present leader is GPT 5.1 Codex Max at more than 30 minutes of non-supervised work. The previous version of Gemini (2.5) was able to work for ~10 minutes, so I expect that Gemini 3.0 will be near the leading edge in the coming weeks.
Progress continues for each of the major AI companies, and the developments in 2025 alone are remarkable. Gemini, with Google’s data-center and intellectual heft, is clearly well-positioned despite it’s earlier hallucinations and AI issues. OpenAI is reportedly alarmed by these tools and have declared “code red” to improve their product quickly.
Below, you’ll find some of the more interesting examples I’ve seen of Gemini / Nano Banana Pro.
What does this mean? Are you a student struggling to understand your math assignment? Simply snap a photo of your assignment, and Gemini will complete the work for you. This could be helpful for understanding and checking work. But this could also be a very easy way to cheat, as I’ve read some reports that Gemini is able to replicate an individual’s handwriting style (meaning someone wouldn’t have to manually copy the work). For instructors, the idea of a student mastering take-home assignments may no longer be a viable indicator of learning.
I asked Nano Banana Pro to generate a graphic novel telling the story and explaining the most important concepts based on a summary I provided. Here is the result:
Grigory’s post is fun and features a variety of comics with differing styles that communicate core ideas of very technical research. This, of course, won’t supplant the importance of original research, but it has the possibility of translating technical knowledge to the masses.
Google also provided some inspiration on ways to use Nano Banana Pro: (Prompt: Create an infographic about this plant focusing on interesting information.)
The prompt and image input are simple, and the generated infographic is interesting and full of helpful details. And best of all, it took a human very little time to create.
What does this mean? Are you in the business of communicating interesting but ultimately difficult-to-understand research? You can generate visuals to convey key ideas within complex research to media and students. Are you in the business of creating infographics? I suspect that if you’re exceptionally talented, you’ll continue your work without much interruption. But for middling designers, your business may dry up as people find more cost-effective ways to create supporting graphics.
Have you used Gemini 3.0 Pro yet? What are your observations of where the tool exceeds?
Leave a Reply