Phone-use agents

APK Arena

Repository Write-up

Levels: 49
Models: 21
Best: 81.3%; gpt-6-astra
Field avg: 58.5%

Overall score

15 of 21 models · mean across every level scored

Cost frontier

15 of 21 models · upper left is better

Leaderboard

15 of 21 models · sortable category breakdown

#
01	gpt-6-astra	81.3	89	96	67	96	73	55	87	51/49
02	claude-opus-5	78.5	89	86	66	94	68	57	88	51/49
03	claude-fable-5.1	75.8	91	86	66	90	64	52	77	50/49
04	gpt-5.6-sol	71.7	82	84	59	86	66	57	72	51/49
05	claude-fable-5	67.6	82	77	67	83	59	39	80	51/49
06	gemini-3.8-flash	64.9	85	83	58	83	41	40	67	51/49
07	gpt-5.6-terra	63.8	84	76	53	84	47	47	54	50/49
08	gpt-5.5	60.8	86	81	51	77	40	35	63	51/49
09	gemini-3.5-flash	58.5	78	69	57	75	38	39	63	51/49
10	gpt-5.6-luna (xhigh)	58.3	79	76	48	79	50	27	45	50/49
11	gemini-3.7-flash	58.1	71	80	53	69	45	36	55	51/49
12	claude-opus-4.8	57.9	74	71	45	68	48	38	61	50/49
13	muse-spark-1.3	56.2	77	81	38	76	28	31	58	49/49
14	claude-sonnet-5	54.5	68	78	50	71	43	22	57	51/49
15	grok-4.6	50.9	61	73	28	51	45	37	66	51/49

Score by release date

15 of 21 models · dashed line is the trend in best-known score