Model matchups.
Two models, the same four tasks, the same prompts, scored by fixed checks. We only show results from real runs, with every output in full. Where we haven't run a matchup yet, we say so.
All matchups
Claude Sonnet 5.5 vs Claude Haiku 5.5
For everyday tasks, do you need the mid-priced model, or does the cheapest one do as well?
Pilot run
Sonnet 5.5 26/26 · Haiku 5.5 26/26 checks
Claude Sonnet 5.5 vs GPT-6.1 Sol
Mid-range models from the two biggest providers, on the same everyday tasks.
Not yet run
Claude Sonnet 5.5 vs Gemini 3.8 Flash
Two fast, lower-cost models, on the same everyday tasks.
Not yet run
The test suite, version 1.0
Coding
Write a small parsing function
Small, well-specified functions are the everyday coding job. Hidden tests decide the score, including invalid input the function must reject.
11 automatic checks
Research
Answer only from a source
Research with AI fails when the model fills gaps from its own guesses. One question has no answer in the source, and the right reply is to say so.
6 automatic checks
Writing
Rewrite a notice in plain English
Plain-English rewriting is one of the most common writing jobs. The checks test that every fact survives, the word limit holds and the jargon goes.
7 automatic checks
Reasoning
Solve a scheduling puzzle
A small logic puzzle with exactly one solution, checked by brute force when we wrote it. The score depends on the answer, not the working.
2 automatic checks
Every prompt is shown in full on each matchup page. Prices and specs come from our dated model profiles.