Sol, Sonnet, Argon: comparing the new AI models

The short answer

An announced model is not necessarily a usable alternative. Our overview combines provider information with a plan for your own comparison. It does not claim a winner based on our own measurements.

Your practical comparison

What you need

A real task, an expected outcome and access to the comparison models. Use anonymized data and record model name, date, reasoning setting and tools.

1. Choose three tasks: a known code defect, a structured analysis and a writing assignment with clear requirements.
2. Give each model the same files and requirements. Set a time limit and correction budget in advance.
3. Check outcomes: does it work, are the facts correct and were all requirements followed?
4. Record waiting time, your own rework and actual costs separately.
5. Repeat difficult tasks before deciding which model should handle them in future.

Example to get you started

Test prompt: “Group these anonymized comments into five themes. Include a supporting passage for each and mark missing information.” Criteria: correct grouping, no invented quotations and clearly identified gaps.

Things to consider

The table describes access and useful evaluation tasks. Our suggestions are editorial analysis. Different provider benchmarks do not form a common leaderboard. The creator videos below present outside experiences; titles and conclusions belong to their authors.