The AI models I Use for Work
Published Nov 11, 2025
⋅
2 minutes read
Here are some rough notes on the AI models I use as of November 11th, 2025.
Important Caveats
- The models listed below reflect my current personal preferences and what works best for my specific workflows.
- I strongly encourage everyone to look at benchmarks and then test out different models for themselves.
- This list is a snapshot in time and is likely to change monthly, if not more frequently, as new models are released, existing ones are updated and I continue exploring new ways of working with them.
Search & Answer Engines
- Kagi : Google search, but better, with customizable results and no ads.
- Kagi Assistant with GLM-4.6 : Google's AI answer mode, but better, with my own custom prompt and AI model selection.
- Gemini Advanced Deep Research : For detailed but well organized reports that give me a lot to think about.
Writing & Content Creation
- Claude Sonnet 4.5 : My go-to.
- Gemini 2.5 Pro : My occasional second option while we wait for Gemini 3.
- Kimi K2 Thinking & GLM-4.6 : New and impressive models I'm exploring through Ollama Cloud and slightly prefer to Claude and Gemini, so far.
- Granite-4-3b: An LLM I can run on my laptop via Ollama with total privacy and without needing an internet connection.
Image Generation
- Midjourney : My go-to, and it's not even close.
Utility Tasks
- Gemini-2.5 Flash : Great for quick tasks like reformatting text, reorganizing information and other minimally complex data manipulation.
- Claude Sonnet 4.5 (Extended Thinking) : My choice for summarizing lengthy texts, synthesizing research from multiple documents, and structuring longer documents.
Powering AI Helpers
I often use Claude Sonnet 4.5 and Gemini 2.5 Pro together to power my AI collaborators. Each model has its own quirks, but using them together helps surface more insights and perspectives.
Claude is very blunt, Gemini is too nice, and I find GPT-5's default writing style is so annoying that I don't want to use it at all. Supposedly GPT-5.1 fixes this.
Coding
- v0 : My vibe-coding go-to.
- Cursor : The best code autocomplete I've found for React / Typescript. Their new model, Composer-1, is a joy to work with in Agent mode.
- Claude Code : For more complex tasks, Claude code with Sonnet 4.5 has completely replaced my use of Cursor's coding agent. The new UI, planning mode and higher usage limits make it a no-brainer for React / Typescript codebases like mine.
Feel free to use this list as a starting point, but please experiment and discover what works best for you!
PS: The next cohort of my AI literacy for business course starts on November 19th in case you'd like structured guidance on implementing AI tools to support your work.