What the perfect model actually looks like

Three good things pull against each other: smarter, faster, cheaper. Two go on the axes; the third becomes colour (green = better). Pick what matters and read the leaderboard — there is rarely a single winner.

Colour

30 of 50 models have both selected axes. Hover for detail. Ringed dots are on the Pareto frontier; the ★ is your current best pick.

Intelligence (higher is better)

GPT-5.5 (xhigh)

★GPT-5.5 (high)

GPT-5.4

GPT-5.4 mini

gpt-oss-120b

Gemini 3.5 Flash

Gemini 3.1 Flash-Lite

Gemini 3.1 Pro

Gemma 4 31B

Claude Opus 4.8

Claude Opus 4.7

Claude Sonnet 4.6

Claude 4.5 Haiku

DeepSeek V4 Pro

DeepSeek V4 Flash

Qwen3.7 Max

Qwen3.7 Plus

Qwen3.6 Plus

Qwen3.6 35B

Qwen3.5 397B

GLM-5.2

GLM-5.1

Kimi K2.6

Kimi K2.7 Code

MiniMax-M3

MiniMax-M2.7

MiMo-V2.5-Pro

Grok 4.3

Nemotron 3 Ultra

Mistral Medium 3.5

Time / task (lower is better)

worsecolour = cost / task (grey = no data)better

Intelligence55%

Speed15%

Low cost30%

These weights rank the 30 models with complete intelligence + time + cost. The chart axes above are independent.

#	Model	Intel	Time	$/task	tok/s	Score
1	GPT-5.5 (high)OpenAIfrontier	53.1	3.36m	$0.72	59	66
2	GPT-5.5 (xhigh)OpenAIfrontier	54.8	4.81m	$0.83	58	66
3	DeepSeek V4 ProDeepSeekfrontier	44.3	8.23m	$0.0484	75	64
4	DeepSeek V4 FlashDeepSeekfrontier	40.3	6.78m	$0.0279	107	63
5	Gemini 3.1 ProGooglefrontier	46.5	1.74m	$0.3372	123	63
6	Gemini 3.5 FlashGooglefrontier	50.2	2.91m	$0.6811	161	63
7	GLM-5.2Z AIfrontier	51.1	6.01m	$0.4645	102	63
8	MiMo-V2.5-ProXiaomifrontier	42.2	8.88m	$0.0322	45	62
9	GPT-5.4OpenAI	51.4	3.87m	$0.99	150	61
10	Claude Opus 4.8Anthropicfrontier	55.7	6.95m	$2.05	62	59
11	MiniMax-M3MiniMaxfrontier	44.4	6.51m	$0.1567	62	59
12	Qwen3.7 PlusAlibabafrontier	39	6.63m	$0.0492	50	58
13	Claude Opus 4.7Anthropic	53.5	6.2m	$1.97	—	57
14	Qwen3.7 MaxAlibaba	46	3.33m	$0.68	95	56
15	Kimi K2.7 CodeKimifrontier	41.9	5.25m	$0.1838	54	56
16	Grok 4.3xAIfrontier	37.6	1.44m	$0.1858	152	54
17	MiniMax-M2.7MiniMax	38.1	7.25m	$0.0742	44	53
18	Nemotron 3 UltraNVIDIAfrontier	37.8	2.36m	$0.2446	170	51
19	GLM-5.1Z AI	40.2	5.57m	$0.2404	82	51
20	Qwen3.6 PlusAlibaba	39.6	6.4m	$0.2667	—	48
21	Kimi K2.6Kimi	42.8	11.68m	$0.3146	44	45
22	Gemini 3.1 Flash-LiteGooglefrontier*	25	1.14m	$0.043	284	44
23	Qwen3.6 35BAlibabafrontier	31.6	2.64m	$0.1784	175	43
24	GPT-5.4 miniOpenAI	40	7.8m	$0.5048	173	43
25	Claude Sonnet 4.6Anthropic	47.2	13.2m	$1.14	58	42
26	Gemma 4 31BGooglefrontier	29.4	5.57m	$0.08	34	41
27	Qwen3.5 397BAlibaba	33.7	5.15m	$0.3331	51	39
28	gpt-oss-120bOpenAI	23.8	1.75m	$0.0607	338	39
29	Claude 4.5 HaikuAnthropic	31	2.79m	$0.33	110	38
30	Mistral Medium 3.5Mistral	29.9	5.23m	$0.5971	81	30
Incomplete on Artificial Analysis (not ranked)
—	Claude Fable 5Anthropic	60	—	$3.25	—	—
—	GPT-5.5 (medium)OpenAI	47	—	—	—	—
—	Gemini 3.5 Flash (medium)Google	45	—	—	159	—
—	Muse SparkMuse	43	—	—	—	—
—	DeepSeek V4 Pro (high)DeepSeek	41	—	—	65	—
—	MiMo-V2-ProXiaomi	40	—	—	—	—
—	GLM-5Z AI	40	—	—	75	—
—	GLM-5-TurboZ AI	38	—	—	—	—
—	Kimi K2.5Kimi	38	—	—	53	—
—	DeepSeek V4 Flash (high)DeepSeek	37	—	—	—	—
—	GLM-5V-TurboZ AI	35	—	—	—	—
—	MiniMax-M2.5MiniMax	34	—	—	186	—
—	Hunyuan 3Tencent	34	—	—	119	—
—	DeepSeek V4 Pro (non-reasoning)DeepSeek	31	—	—	76	—
—	GPT-5 mini (medium)OpenAI	31	—	—	100	—
—	Grok 4.1 FastxAI	31	—	—	—	—
—	DeepSeek V4 Flash (non-reasoning)DeepSeek	29	—	—	101	—
—	Trinity LargeTrinity	25	—	—	160	—
—	Gemini 3 FlashGoogle	—	—	—	169	—
—	Grok 4.3 (medium)xAI	—	—	—	155	—

All metrics per task, from Artificial Analysis (19 Jun 2026): Intelligence Index, time per task (min), cost per task (USD, AA’s official figure), output speed (tok/s) and output tokens per task. AA’s scrape caps each chart at 20 rows, so values for higher-cost/slower models (Opus, Fable, Sonnet, GPT-5.5 tiers) are read from AA’s own cost/time charts and marked *. A model is listed as incomplete only where AA shows no value at all. The Pareto frontier and ranking use the 30 fully-characterised models. Figures move — refresh modelData.ts before quoting.