Phone-use agents

APK Arena

Repository Write-up

Levels: 51
Models: 32
Best: 81.4%; gpt-6-astra (high)
Field avg: 62.1%

Overall score

15 of 32 models · mean across every level scored

Cost frontier

15 of 32 models · upper left is better

Leaderboard

15 of 32 models · sortable category breakdown

#
01	gpt-6-astra (high)	81.4	89	95	73	95	72	61	87	51/51
02	claude-opus-5.5 (high)	81.3	90	93	66	96	75	62	85	51/51
03	gpt-6.1-sol (high)	78.9	87	90	64	96	75	57	83	51/51
04	claude-opus-5 (high)	78.5	89	86	66	94	68	60	88	51/51
05	gemini-4-argon (low)	76.0	82	93	71	92	65	55	77	51/51
06	claude-fable-5.1 (high)	75.8	91	86	66	90	64	56	77	50/51
07	claude-sonnet-5.5	72.3	79	88	68	87	69	44	76	51/51
08	gpt-5.6-sol (high)	71.7	82	84	59	86	66	52	72	51/51
09	gpt-6-sol (high)	70.9	84	82	70	86	51	56	70	51/51
10	claude-fable-5 (high)	67.6	82	77	67	83	59	34	80	51/51
11	muse-wonder-1.0 (high)	67.2	74	85	52	80	66	44	68	51/51
12	claude-haiku-5.5 (high)	65.0	72	85	48	76	56	42	78	51/51
13	gemini-3.8-flash (high)	64.9	85	83	58	83	41	40	67	51/51
14	gpt-5.6-terra (high)	63.8	84	76	53	84	47	45	54	50/51
15	gpt-5.5 (high)	60.8	86	81	51	77	40	30	63	51/51

Score by release date

15 of 32 models · dashed line is the trend in best-known score