Which AI is good at what, and what it costs, across thirteen kinds of work. Select a model or vendor to mark it on every chart. Every number links to its source; every benchmark shows how it is run.
Same model, more thinking. Score at each effort level, low to max.
What one task costs each model at its highest tested effort. Log scale; left is cheaper. Agentic tasks cost 10 to 1000 times more than a Q&A.
How each product line moved from one generation to the next on the same benchmark (best effort of each).
Best-effort score of each model on each benchmark. Colour = rank within that column (darker = better). Click a model for its page, a column for the benchmark.
Every point in the data file carries its exact source URL and the fragment it was read from.