The best LLMs for different uses.

As an avid AI enthusiast, I thought I’d list out all the different models I use for different use cases.

Please note that the field moves quickly and this list may become outdated quickly.

Reach out if you ever have a question about what I’d recommend!

Creative writing:
1. According to the Creative Writing EQ Bench, Polaris alpha (aka GPT-5.1, not sure why they haven’t updated the name) tops the benchmarks. I agree with this, however I also really like Kimi-K2’s style. It’s different from other LLM outputs, and usually swap between the two models.
  1. TLDR: Use normal ChatGPT with thinking OR use Kimi K2 for creative writing.
  2. Benchmark:
Maths:
1. Honestly all frontier models are more than adequate for 99.9% of math use cases. That being said, Gemini 3.0 pro currently is at the top of math benchmarks.
2. Benchmark:
Search/Research:
1. For fast search: Grok 4.1 fast or ChatGPT with web search enabled. I’ve found both these models to be fast and surface relevant results.
2. For deep research: I use a combination of Deep Research on ChatGPT and Gemini, and also sometimes Exa Research. (Exa is doing pretty cool things with regards to building infrastructure for AI models, check them out!)
Coding:
1. I use a combination of the recently released Opus 4.5, and Gemini 3.0 Pro. Cursor with the student plan is super generous with token and rate limits, I’d highly recommend anyone who hasn’t coded before/is curious about coding to give Cursor a try. Reach out if you have any questions!
Image generation:
1. Nano-Banana Pro (aka Gemini 3 Pro image) is by far the best option currently. Only available via Gemini, it significantly outperforms everything else and is super useful for creating flowcharts/diagrams for learning!
  1. “Make an infographic showing like the differences between AIN and ATN and prerenal vs intrarenal vs posteenal AKIs”
  2. “Create an infographic to show me how binary trees work”
Video generation:
1. I believe Veo 3.1 or Sora 2 are SOTA for this, not really experimented too much with video models. Take this with a grain of salt.
Presentation creation:
1. Highly recommend Kimi-K2, they have a purpose built harness for slide generation which calls Nano Banana Pro for images. Check it out here!
OCR / Handwriting to text:
1. OlmoOCR by AllenAI or one of the Qwen-3-VL models are state of the art (SOTA) for this. I recommend checking out AllenAI they have some pretty cool open source models.
Speech to text / Transcription:
1. Highly recommend NVIDIA’s Parakeet models for local use, they’re extremely fast and highly accurate. If looking for an API, Qwen-ASR is SOTA and I’m really looking forward to them opensourcing it.

Links: