Short answer
For a question in plain English, semble is better: it puts the right file in its top 10 for 100% of 45 queries, Pixel for 69%, and it costs less context. Pixel adds the call graph, blast radius, Git history and guarded Git writes. They combine: semble for search and Pixel for the rest cost about 5,100 tokens of context per turn.
Side by side
Measured head to head. Pixel and semble on the same cases, on the same machine.
| Measure | Pixel | semble |
|---|---|---|
| Right file in the top 10 (45 plain-English queries) | 69% | 100% |
| Context on every turn | ~4,160 tok | ~980 tok |
| Call graph, blast radius | Yes | No |
| Git history, Git writes | Yes | No |
Every figure above is on the benchmarks page, with its sample size, its method and the cases where Pixel loses.
Where semble wins
Natural-language search, and a lighter context on every turn.
How they differ
semble is a search tool for questions in plain English, and it does that job better than pixel search-meaning. Pixel is a wider index: pixel who-calls, pixel impact and pixel scope-task answer from the call graph, and its history and Git commands cover what a search tool does not.
Using both
They combine. Route plain-English search to semble and the structural questions to Pixel; the short answer above gives what the pair costs in context.