Model benchmark rankings

Compare scores, test time and reference cost across models and reasoning levels.

Tested on our private datasetsWhy we test this way

Updated 9/30/2026

gpt-6.1-sol

NewWorth a look

93.8Overall

claude-sonnet-5-5

New

85.8Overall

claude-opus-5-5

New

89.1Overall

Each model shows its best score here. Expand to see other effort levels.33 models · 83 configurations

About the scores

Backend 40% · Frontend 30% · Reasoning 30%, from the same tested configuration. Each model shows its highest-scoring tested effort for the selected category.

Time and cost follow the selected category. They describe these evaluations, not the speed or price of a typical request.

Explore the tasks →About the method →

ⓘ

Rank by	Model / Tested effort		Backend	Frontend	Reasoning	Test time	Reference cost
01	gpt-6-astra xhigh	95.3	95	96	95	83m 56s	$7.01
02	gpt-6.1-solNEW max	93.8	95	96	90	62m 9s	$1.69
03	claude-opus-5-5NEW max	89.1	81	89	100	33m 57s	$3.67
04	claude-fable-5-1 max	89.0	83	91	95	79m 7s	$26.15
05	claude-sonnet-5-5NEW xhighAnthropic	85.8	84	89	85	48m 32s	$2.80
06	grok-4.7 high	84.3	87	80	85	196m 18s	$5.25
07	gpt-6-solNEW xhigh	81.5	77	94	75	39m 22s	$1.10
08	claude-opus-5 max	81.0	66	87	95	92m 12s	$7.99
09	grok-4.6 xhigh	79.5	75	80	85	144m 16s	$2.51
10	gpt-5.6-sol max	77.6	83	73	75	87m 21s	$3.25
11	claude-opus-4-8 max	75.9	60	88	85	106m 52s	$16.33
12	k3 highMoonshot	72.5	65	85	70	158m 17s	$3.28
13	mimo-v2.6-proNEW defaultXiaomi MiMo	71.4	66	85	65	336m 3s	$0.66
14	gpt-5.6-terra max	70.5	78	76	55	133m 2s	$1.94
15	glm-5.3 max	70.0	64	78	70	156m 42s	$1.91
16	space-bunnyNEW defaultOther	69.3	57	80	75	62m 35s	$0.000
17	hy4-preview highTencent Hunyuan	69.0	69	73	65	218m 41s	$1.63
18	deepseek-v4.1-flash max	68.7	57	88	65	35m 40s	$0.41
19	MiniMax-M3.1-Flash-PreviewNEW max	68.0	71	57	75	54m 7s	N/ANo cost data
20	Qwen 3.8 Flash defaultQwen	67.1	65	87	50	134m 31s	$0.42
21	step-5-previewNEW defaultStepFun	66.2	56	71	75	98m 22s	$1.24
22	k2.8 preview high	64.2	54	72	70	70m 58s	$0.66
23	deepseek-v4-pro max	64.0	64	63	65	93m 15s	$1.15
24	Gemini 3.8 Flash High defaultGoogle	61.8	60	86	40	25m 21s	$0.13Partial cost
24	gpt-6-lunaNEW xhigh	61.8	57	80	50	63m 52s	$0.061
26	deepseek-v4-flash high	60.5	53	76	55	98m 25s	$0.42
27	deepseek-v4-flash-vision-exp high	56.7	57	43	70	88m 43s	$0.32
27	Qwen 3.8 Max defaultQwen	56.7	60	69	40	133m 3s	$3.41Partial cost
29	mimo-v2.6-flashNEW defaultXiaomi MiMo	56.4	54	71	45	386m 14s	$0.28
30	gpt-5.6-luna max	55.3	61	78	25	86m 14s	$0.35
31	Gemini 3.7 Flash High defaultGoogle	52.4	59	71	25	24m 19s	$0.13
32	glm-5.3-flash max	51.7	43	75	40	96m 41s	$0.074Partial cost
32	MiniMax-M3 highMiniMax	51.7	43	65	50	56m 33s	$0.62Partial cost

Each model uses its highest-scoring effort here. Equal scores share a rank.

Backend 40% · Frontend 30% · Reasoning 30%, from the same tested configuration. Evaluation scope and sources →