Worksheet: meet the models (sample, filled in)
An example of what a completed worksheet might look like. Your numbers and observations will differ, that's fine; the point is the comparison, not matching this. (Specs change often; always check the maker's current pricing page.)
The two models I compared
| Model A | Model B | |
|---|---|---|
| Name & maker | Claude (Anthropic) | ChatGPT / GPT (OpenAI) |
| Step 2, tone | warm, plain; stuck to 3 sentences | friendly; ran slightly long (4 sentences) |
| Step 2, really 3 sentences? | yes | not quite |
| Step 3, recent-question result | said it couldn't be sure of the latest result (admitted a cutoff) | gave a confident answer that was out of date |
| Step 4, context window | large (hundreds of thousands of tokens) | large (varies by model) |
| Step 4, price (per 1M tokens) | cheaper "small" model vs pricier "frontier" one | same pattern: a budget model + a frontier model |
Step 5: which would I choose?
- Free, quick personal chat: either free web tier, pick whichever I find clearer; cost isn't a factor for occasional personal use.
- Must keep all data on my own computer: neither hosted model, I'd use an open-source model run locally (e.g. via Ollama/LM Studio), because hosted models send data off my machine.
One thing that surprised me
One model admitted it didn't know a recent result while the other made one up confidently, a vivid example of why you can't trust an answer just because it sounds sure (→ RAG + guardrails).