Models, compared
Same question, different answers, from the numbers each lab publishes.
-
Opus vs GPT vs Gemini: what the benchmarks actually say
The three labs' own tables disagree, because each one reports the tests it wins. The prices differ fivefold and the index scores differ by two points.
-
Open models you can run on one machine
Three open-weight families fit on a single Mac or a single GPU. What decides whether they run is total parameters, not active ones.
-
Chinese models: Qwen, DeepSeek and Kimi compared
DeepSeek sells throughput, Qwen sells breadth, Kimi sells long-context coding. The three flagships cost between $0.96 and $2 per million input tokens.