AI Productivity
Pick the model for the job
There isn't one best AI model. Some are better at generating images, some are better at structured data work, some are cheaper to run at volume. Inside a company, the team should still share one agentic platform. Outside that, it's worth testing other models to see what works best for each job.
What people like about each
A rough guide, not a leaderboard. Models change every few months, so treat this as a starting point for your own testing.
| Model family | What people like it for | Worth knowing |
|---|---|---|
| Claude (Anthropic) | Long documents, careful writing and editing, and agentic coding in tools like Claude Code. | My default for day-to-day product work. |
| GPT (OpenAI) | Image generation (ChatGPT Images), including readable text inside images, and structured outputs that stick to a JSON schema you define. | Strong pick when an image or a reliably shaped data file is the deliverable. |
| Qwen (Alibaba) | Open-weight models in every size from phone to data centre, vision models that read screens and documents, and low API prices. | Apache 2.0 licence, so you can self-host it or fine-tune it. |
| Kimi (Moonshot AI) | Agentic coding and tool use over long sessions, with a 256K-token context window. | Open weights; recent releases focus on using fewer tokens per task. |
Match the model to the job
- Images. If the output is a picture, such as a mock-up, a social graphic or a diagram with labels, use a model built for image generation. General chat models vary a lot here, especially with text inside the image.
- Structured data. If the output feeds a spreadsheet, a database or another tool, pick a model that can be held to a schema, and check a sample of rows by hand before you trust the rest.
- Volume. For high-volume, low-stakes work like tagging, reformatting or short summaries, a smaller open-weight model is often good enough and costs a fraction of the price.
- Long, multi-step work. For anything that runs across many files or tools, stick with a model known for agentic work.
In a company, share one platform
When you're working in a company, it's better for everyone to work on the same agentic platform, such as Claude Code. Skills, workflows and prompts only compound when the whole team can share and reuse them, and that breaks down once people are spread across different tools.
Outside that, try other models
For your own experiments, and for jobs that sit outside the team's shared workflows, test other models. Take one real task you do every week, run the same prompt through two or three models, and compare the results for quality, speed and cost. You'll usually find one clear winner for that job, and it isn't always the model you use for everything else. A platform that gives you several models behind one account, like TensorX, makes this quick to do.
Keep a short note of what worked for which job. That note becomes your own model guide, and it'll be more useful than any benchmark because it's based on your own work.