Independent AI evaluations
Less hype.
More evidence.
We test AI models, agents, and developer tools.
Independent research. Open methods. Results you can verify.
Latest from the lab 02
Research & perspectives- benchmarkModels
Do AI models recommend their own company's products? We asked 10 models 1,680 times
Kimi, Claude, GPT and Amazon Nova put their maker's products first more often than other models do. Claude names Anthropic first in 71% of LLM API answers, versus 20% for other models. Gemini under-ranks Google Search.
- essayAbout the lab
Why we run our own benchmarks
Vendor benchmarks are run by the vendor, on the vendor's tasks, at launch. We re-run comparisons independently, publish the raw data, and update results as models change.
01 / Measured
Research
Benchmarks, comparisons, and field trials. Exact versions, real costs, and the data behind every result.
Explore research02 / Considered
Essays
Ideas and opinions shaped by hands-on testing. What the numbers mean, and what they leave out.
Read the essays