Quality ranking · LMArena Elo

Local LLM quality leaderboard

Open models you can run locally, ranked by LMArena text Elo as of 2026-06-15. Higher is better. Only models present on the arena are listed, so many small and distilled models are absent. Scores drift over time.

Model 16 entries

#	Model	Params	Elo
1	Qwen3 30B-A3B MoE	30.5B	1383
2	Gemma 3 27B	27B	1366
3	Mistral Small 3 24B	24B	1357
4	Qwen3 32B	32B	1347
5	Gemma 3 12B	12B	1342
6	Llama 3.3 70B	70B	1318
7	Gemma 3 4B	4B	1303
8	Qwen2.5 72B	72B	1303
9	Gemma 2 27B	27B	1289
10	Gemma 2 9B	9B	1266
11	Phi-4 14B	14B	1256
12	Llama 3.1 8B	8B	1211
13	Llama 3.2 3B	3B	1166
14	Mistral 7B	7B	1149
15	SmolLM2 1.7B	1.7B	1114
16	Llama 3.2 1B	1B	1110

Source: LMArena text leaderboard, snapshot around 2026-06-15. Elo measures human preference in blind pairwise chats, not local speed. Pick the highest-Elo model your hardware can run on the device pages.