EMAILS
Decided by people's blind votes on LMArena (instruction-following, 19 Nov) until our own blind test has enough votes. Full ranking
No benchmark measures how well an email reads; the blind-test votes are the real answer.
Four quick questions. You get Keep, Drop or Switch for each plan you pay for, the best plan at each price, and a line for each task, each with its source. Nothing you tap leaves your device.
This compares answers only. If you use an app for something else (voice, files, images, research), that may be worth keeping it for.
Decided by people's blind votes on LMArena (instruction-following, 19 Nov) until our own blind test has enough votes. Full ranking
No benchmark measures how well an email reads; the blind-test votes are the real answer.
Decided by people's blind votes on LMArena (coding and WebDev, 20 Nov) until our own blind test has enough votes. Full ranking
LMArena's coding and WebDev boards show which code people preferred, not whether it runs.
Decided by people's blind votes on LMArena (maths, 19 Nov) until our own blind test has enough votes. Full ranking
These test hard questions; school-level maths no longer separates the top models.
Decided by people's blind votes on LMArena (instruction-following, 19 Nov) until our own blind test has enough votes. Full ranking
Short English documents only; nothing yet for long documents.
Decided by people's blind votes on LMArena (overall, 19 Nov) until our own blind test has enough votes. Full ranking
Decided by people's blind votes on LMArena (overall, 19 Nov) until our own blind test has enough votes. Full ranking
Decided by people's blind votes on LMArena (coding, 19 Nov) until our own blind test has enough votes. Full ranking
No benchmark for spreadsheets; LMArena's coding ranking stands in.
Decided by people's blind votes on LMArena (creative-writing, 19 Nov) until our own blind test has enough votes. Full ranking
No current creative-writing benchmark besides LMArena.
Decided by people's blind votes on LMArena (Spanish and German, 19 Nov) until our own blind test has enough votes. Full ranking
Spanish and German only.
Decided by people's blind votes on LMArena (multi-turn, 19 Nov) until our own blind test has enough votes. Full ranking
This check covers ChatGPT, Claude, Gemini, DeepSeek and Kimi, so other apps are left out of it.
Nothing you tapped left your device: your answers are only in this page's address, after the "#". US prices, before tax, checked Sat 7 Nov (plans in detail). Task picks: our blind test, or LMArena's text leaderboards (Thu 19 Nov) and its WebDev leaderboard for coding (Fri 20 Nov) while a task has too few votes. Benchmarks: Epoch AI, CC BY 4.0 (runs up to Fri 13 Nov); Vectara's hallucination leaderboard (Thu 22 Oct). Benchmark scores are at each model's strongest setting, which an app doesn't always use. "Among the best" means level with the leader: their 95% ranges overlap. How we score