Glossary · term113 of 202
mixed-model benchmark
A test comparing combinations of different models working together against a single model working alone on the same task, used to judge whether splitting work across models pays off.
also written as mixed model benchmarks · multi-model benchmark
Related terms
execution subagentA subagent given a narrow task with a fixed output shape and a written boundary, as opposed to open-ended reasoning or planning work. Its job is to follow the spec, not to expand scope.view termfan-outThe pattern where one delegated task spawns multiple concurrent subagents, and each of those can spawn more of its own. Depth is usually capped, but breadth often is not, so total agents alive can multiply far past what anyone intended.view term
At a glance
previousmirrored networking modenextmodel tier- cited by
- 0 posts
- categories
- 0
- first used
- —