Choose a model
There is no best model, only a best model for something. A voice agent needs an answer to start within a second and can live with less intelligence; a research assistant can think for a minute and cannot. Start from the job below, or set your own constraints.
Data as of 21 Oct 2026
Start from the job
Each of these ranks on what that work actually needs, and prints the weighting so you can argue with it.
Voice agents
Which model should I use for a voice agent?
Weighted on time to first answer 45%, output speed 20%, tool use 15%.
All 65 ranked for this →
Coding agents
Which model should I use for an autonomous coding agent?
Weighted on coding 35%, tool use 30%, long context 20%.
All 209 ranked for this →
High-volume extraction
Which model should I use to classify or extract at scale?
Weighted on price per 1m 45%, output speed 25%, instruction following 20%.
All 212 ranked for this →
Deep research
Which model should I use for long analytical work?
Weighted on reasoning 30%, quality 25%, knowledge 25%.
All 381 ranked for this →Or work it out yourself
Find a model
None of those quite your job? Set your own constraints — a budget, a wait you can live with, a licence, a context window — and rank what survives on the one skill you care about.
What more thinking costs
Once you have picked a model, its reasoning setting is the next decision — and it is a price dial as much as a quality one. How much each step up costs, and what it buys.