All 14 models, ranked by real measured speed on 3 coding tasks (explaining a block of code, explaining a system design, refactoring a function). Raw speed alone isn't the whole story: the fastest model for a task this size can be more model than the task needs, or not enough. This site's own fit judgment is shown alongside every ranking.
The literal-fastest model across these tasks is usually GPT-5.6 Luna, but it's built for quick, everyday answers, not reasoning through code. For coding specifically, this site's own fit judgment points to Sonnet 5 (Anthropic) or GPT-5.6 Terra (OpenAI): the models actually sized for understanding how code fits together, not just typing fast.
See the plain-language version: Which model should I use for coding? →
Every number above is a real computed figure: recorded runs where they exist, a clearly-labelled placeholder profile otherwise, same as every result on this site. See how we measure → or the full model comparison → across all 25 tasks.