AI Model Rankings

Raw Firepower

Maximum-Token Performance

Scores are normalized within each eligible benchmark and aggregated only from published results. The axis is zoomed to the displayed range; coverage remains visible because missing benchmarks are not estimated.


1	Claude Fable 5Anthropic	81.14	11,558.5	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
2	GPT-5.6 SolOpenai	80.44	7,342	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
3	GPT-5.5Openai	78.49	7,580	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
4	Claude Opus 4.7Anthropic	77.89	7,019	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
5	Claude Opus 4.6Anthropic	77.07	6,129	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
6	GPT-5.4Openai	76.54	10,763.5	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
7	Claude Sonnet 4.6Anthropic	74.14	7,875.5	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
8	Grok 4.5Xai	73.80	6,602.5	ARC-AGI-1 ARC-AGI-2 LiveBench+6 more 82% ranking coverage
9	GPT-5.2Openai	70.28	7,889	ARC-AGI-1 ARC-AGI-2 BFCL+6 more 82% ranking coverage
10	GPT-5.6 TerraOpenai	69.27	12,533	ARC-AGI-1 ARC-AGI-2 LiveBench+5 more 73% ranking coverage
11	Claude Opus 4.8Anthropic	68.28	13,456	ARC-AGI-1 ARC-AGI-2 LiveBench+5 more 73% ranking coverage
12	Gemini 3.5 FlashGoogle Deepmind	65.91	17,067	ARC-AGI-1 ARC-AGI-2 LiveBench+5 more 73% ranking coverage
13	GPT-5.6 LunaOpenai	64.99	12,547	ARC-AGI-1 ARC-AGI-2 LiveBench+5 more 73% ranking coverage
14	Gemini 3.1 ProGoogle Deepmind	59.90	11,694.5	LiveBench Agent Arena Document Arena+4 more 64% ranking coverage
15	Gemini 3.6 FlashGoogle Deepmind	56.22	12,837.5	ARC-AGI-1 ARC-AGI-2 LiveBench+4 more 64% ranking coverage
16	Claude Opus 5Anthropic	53.89	8,727	ARC-AGI-1 ARC-AGI-2 LiveBench+3 more 55% ranking coverage
17	Muse Spark 1.1Meta	51.24	9,844.5	LiveBench Agent Arena Document Arena+3 more 55% ranking coverage
18	Gemini 3.7 FlashGoogle Deepmind	51.05	15,874.5	ARC-AGI-1 ARC-AGI-2 LiveBench+3 more 55% ranking coverage
19	GPT-5.1Openai	50.38	3,970	ARC-AGI-1 ARC-AGI-2 Document Arena+4 more 64% ranking coverage
20	Grok 4.6Xai	49.48	15,864.5	ARC-AGI-1 ARC-AGI-2 LiveBench+3 more 55% ranking coverage
21	GPT-5Openai	49.02	8,777.8	Aider ARC-AGI-1 ARC-AGI-2+4 more 64% ranking coverage
22	Grok 4.3Xai	46.80	8,169	LiveBench Agent Arena Search Arena+3 more 55% ranking coverage
23	Gemini 3.5 Flash-LiteGoogle Deepmind	45.03	6,746	ARC-AGI-1 ARC-AGI-2 LiveBench+4 more 64% ranking coverage
24	Claude Sonnet 4.5Anthropic	44.70	2,441	ARC-AGI-1 ARC-AGI-2 BFCL+4 more 64% ranking coverage
25	Qwen3.8-MaxQwen	43.42	22,485.5	LiveBench Agent Arena Text Arena+2 more 45% ranking coverage
26	Gemini 2.5 ProGoogle Deepmind	42.26	4,521	Document Arena Search Arena Text Arena+2 more 45% ranking coverage
27	Gemini 3 FlashGoogle Deepmind	41.79	—	Agent Arena Document Arena Search Arena+2 more 45% ranking coverage
28	Kimi-K3Moonshot AI	41.09	14,335	ARC-AGI-1 ARC-AGI-2 LiveBench+2 more 45% ranking coverage
29	Grok 4.20 Multi-AgentXai	40.74	—	ARC-AGI-1 ARC-AGI-2 Search Arena+2 more 45% ranking coverage
30	GPT-5.4 MiniOpenai	40.29	25,003.5	ARC-AGI-1 ARC-AGI-2 LiveBench+3 more 55% ranking coverage
31	Claude Opus 4.5Anthropic	39.38	15,203	BFCL LiveBench Document Arena+2 more 45% ranking coverage
32	GPT-5.4 NanoOpenai	36.76	25,005	ARC-AGI-1 ARC-AGI-2 LiveBench+3 more 55% ranking coverage
33	InklingThinking Machines Lab	35.29	5,785	ARC-AGI-1 ARC-AGI-2 Agent Arena+2 more 45% ranking coverage
34	Claude Sonnet 5Anthropic	35.08	—	Agent Arena Document Arena Text Arena+1 more 36% ranking coverage
35	Muse Spark 1.2Meta	34.73	18,700	LiveBench Agent Arena Text Arena+1 more 36% ranking coverage
36	Qwen3.7-PlusQwen	34.16	—	Agent Arena Document Arena Text Arena+1 more 36% ranking coverage
37	Kimi-K2.6Moonshot AI	34.10	27,001	LiveBench Document Arena Text Arena+1 more 36% ranking coverage
38	GPT-5 MiniOpenai	34.01	3,925	ARC-AGI-1 ARC-AGI-2 BFCL+3 more 55% ranking coverage
39	MiniMax-M3MiniMax	33.30	15,530	LiveBench Document Arena Text Arena+1 more 36% ranking coverage
40	Gemma 4 31BGoogle Deepmind	32.36	—	Agent Arena Document Arena Text Arena+1 more 36% ranking coverage
41	Gemini 2.5 FlashGoogle Deepmind	32.21	4,399	BFCL Text Arena Vision Arena+1 more 36% ranking coverage
42	DeepSeek-V4-FlashDeepSeek	30.22	34,434	ARC-AGI-1 ARC-AGI-2 LiveBench+1 more 36% ranking coverage
43	GPT-4.1Openai	28.66	—	Aider ARC-AGI-1 ARC-AGI-2+3 more 55% ranking coverage
44	GPT-5 NanoOpenai	28.60	7,465	ARC-AGI-1 ARC-AGI-2 BFCL+3 more 55% ranking coverage
45	Kimi-K2.5Moonshot AI	26.02	—	Document Arena Text Arena Vision Arena 27% ranking coverage
46	GLM-5.3Z.ai	25.68	62,090	LiveBench Agent Arena Text Arena 27% ranking coverage
47	Qwen3.8-27BQwen	25.63	28,740	LiveBench Text Arena Vision Arena 27% ranking coverage
48	GLM-5.3-FlashZ.ai	25.45	34,707	LiveBench Text Arena Vision Arena 27% ranking coverage
49	Mistral-Medium-3.5-128BMistral AI	24.59	—	Agent Arena Text Arena Vision Arena 27% ranking coverage
50	Qwen3.7 MaxQwen	24.47	10,772.5	LiveBench Agent Arena tonebench 27% ranking coverage
51	Mistral Large 3Mistral AI	24.20	2,056	Text Arena Vision Arena tonebench 27% ranking coverage
52	GPT-4.1 MiniOpenai	24.10	—	Aider BFCL Text Arena+1 more 36% ranking coverage
53	GPT-4oOpenai	23.40	1,174	Aider Text Arena Vision Arena+1 more 36% ranking coverage
54	DeepSeek-V4-Pro-0813DeepSeek	22.95	18,922	ARC-AGI-1 ARC-AGI-2 tonebench 27% ranking coverage
55	Command ACohere	22.65	1,177	Aider BFCL Text Arena+1 more 36% ranking coverage
56	Claude Haiku 4.5Anthropic	21.81	—	ARC-AGI-1 ARC-AGI-2 Document Arena+1 more 36% ranking coverage
57	GPT-4o MiniOpenai	21.36	1,347	Aider Text Arena Vision Arena+1 more 36% ranking coverage
58	GPT-4.1 NanoOpenai	19.78	—	Aider BFCL Text Arena+1 more 36% ranking coverage
59	Llama-4-Scout-17B-16E-InstructMeta	19.70	—	BFCL Text Arena Vision Arena 27% ranking coverage
60	Claude Fable 5.1Anthropic	17.84	—	ARC-AGI-1 ARC-AGI-2 18% ranking coverage
61	ERNIE 5.1Baidu	17.74	—	Search Arena Text Arena 18% ranking coverage
62	Qwen3.5-397B-A17BQwen	17.33	—	Text Arena Vision Arena 18% ranking coverage
63	Gemma 4 26B-A4BGoogle Deepmind	17.28	—	Text Arena Vision Arena 18% ranking coverage
64	GPT-5.5 ProOpenai	17.22	—	ARC-AGI-1 ARC-AGI-2 18% ranking coverage
65	Llama-4-Maverick-17B-128E-InstructMeta	17.19	—	Aider Text Arena Vision Arena 27% ranking coverage
66	GLM-5.2Z.ai	16.94	23,463	LiveBench Text Arena 18% ranking coverage
67	GPT-5.4 ProOpenai	16.91	—	ARC-AGI-1 ARC-AGI-2 18% ranking coverage
68	DeepSeek-V4-ProDeepSeek	16.83	35,014	LiveBench Text Arena 18% ranking coverage
69	Hy3Tencent	16.71	—	Agent Arena Text Arena 18% ranking coverage
70	Kimi-K2-InstructMoonshot AI	15.20	1,617.1	Aider BFCL 18% ranking coverage
71	Grok Build 0.1Xai	14.69	980	LiveBench Agent Arena 18% ranking coverage
72	GPT-5.2 ProOpenai	13.68	—	ARC-AGI-1 ARC-AGI-2 18% ranking coverage
73	Llama-3.3-70B-InstructMeta	12.61	—	BFCL Text Arena 18% ranking coverage
74	gpt-oss-120bOpenai	12.57	3,806.6	Aider Text Arena 18% ranking coverage
75	Nova 2 LiteAmazon	12.40	—	BFCL Text Arena 18% ranking coverage
76	phi-4Microsoft	11.78	—	BFCL Text Arena 18% ranking coverage
77	Gemini 2.5 Flash-LiteGoogle Deepmind	11.33	2,360	BFCL tonebench 18% ranking coverage
78	Llama-3.1-8B-InstructMeta	11.14	—	BFCL Text Arena 18% ranking coverage
79	GLM-5.1Z.ai	8.84	—	Text Arena 9% ranking coverage
80	GLM-5Z.ai	8.74	—	Text Arena 9% ranking coverage
81	DeepSeek-V3.2DeepSeek	8.61	—	Text Arena 9% ranking coverage
82	MiniMax-M2.7MiniMax	8.49	—	Text Arena 9% ranking coverage
83	Qwen3.5-35B-A3BQwen	8.43	—	Text Arena 9% ranking coverage
84	Muse Glimmer 30BMeta	8.41	—	Text Arena 9% ranking coverage
85	DeepSeek-V4-Flash-Vision-ExpDeepSeek	8.37	52,644	LiveBench 9% ranking coverage
86	DeepSeek-R1DeepSeek	8.29	—	Text Arena 9% ranking coverage
87	GPT-5 ProOpenai	8.28	—	ARC-AGI-1 ARC-AGI-2 18% ranking coverage
88	Sonar Reasoning ProPerplexity	8.23	—	Search Arena 9% ranking coverage
89	Granite 4.2 30BIBM	8.23	—	Text Arena 9% ranking coverage
90	MiniMax-M2.5MiniMax	8.21	—	Text Arena 9% ranking coverage
91	Sonar ProPerplexity	8.17	—	Search Arena 9% ranking coverage
92	NVIDIA-Nemotron-3-Nano-30B-A3B-BF16NVIDIA	8.15	—	Text Arena 9% ranking coverage
93	DeepSeek-V3DeepSeek	8.05	—	Text Arena 9% ranking coverage
94	NVIDIA Nemotron 3.5 Lightning 30B-A3BNVIDIA	8.05	—	Text Arena 9% ranking coverage
95	Granite 4.2 8BIBM	7.96	—	Text Arena 9% ranking coverage
96	Olmo-3.1-32B-InstructAi2	7.92	—	Text Arena 9% ranking coverage
97	Olmo-3-32B-ThinkAi2	7.85	—	Text Arena 9% ranking coverage
98	Granite 4.2 3BIBM	7.85	—	Text Arena 9% ranking coverage
99	granite-4.1-8bIBM	7.80	—	Text Arena 9% ranking coverage
100	gpt-oss-20bOpenai	7.78	—	Text Arena 9% ranking coverage
101	Llama-3.1-70B-InstructMeta	7.62	—	Text Arena 9% ranking coverage
102	Gemini 3.1 Flash-LiteGoogle Deepmind	7.28	1,628	tonebench 9% ranking coverage
103	Command A ReasoningCohere	6.42	2,002	tonebench 9% ranking coverage
104	Ministral-8B-Instruct-2410Mistral AI	1.71	—	BFCL 9% ranking coverage

Rank is the table order. Each benchmark leader earns 100 points; other models earn proportional points. Missing results earn no points, so coverage matters. If a model has multiple runs, this table uses its highest published output-token spend.

Value Frontier

LiveBench Quality Versus Estimated Output Cost

LiveBench

Efficiency FrontierLiveBench overall · output estimate

Clear highlight

DeepSeekZ.aiMiniMaxOpenaiGoogle DeepmindQwenMoonshot AIAnthropic

The blue dotted line connects measured, non-dominated observations. Developer-colored dashed lines connect models only when their sourced family and generation match. Cost is estimated from published output tokens and the lowest current USD output rate; it excludes input, caching, batch discounts, and provider-specific benchmark execution details.

Cost Ranking

Score-Adjusted Output Cost

Cost, RankedScore-adjusted output estimate · lowest first

Clear highlight

Cheapest is first. Bars use the same score-adjusted output-cost estimate as the scatter plot; hover details preserve the unadjusted estimate, token count, published rate, and provider.

Token Efficiency

Efficiency Ranking


1	Grok Build 0.1	33.33	82.14	980	LiveBench 33% token benchmark coverage
2	Command A	33.33	68.38	1,177	tonebench 33% token benchmark coverage
3	Kimi-K2-Instruct	33.33	67.17	1,617.1	Aider 33% token benchmark coverage
4	GPT-4o	32.19	65.87	1,174	tonebench 33% token benchmark coverage
5	Gemini 3.1 Flash-Lite	28.21	80.06	1,628	tonebench 33% token benchmark coverage
6	GPT-4o Mini	27.80	65.27	1,347	tonebench 33% token benchmark coverage
7	Gemini 3.5 Flash-Lite	26.70	79.55	6,746	LiveBench tonebench 67% token benchmark coverage
8	Claude Opus 4.7	24.61	94.77	7,019	LiveBench tonebench 67% token benchmark coverage
9	Claude Opus 4.6	24.56	92.74	6,129	LiveBench tonebench 67% token benchmark coverage
10	Claude Opus 5	24.34	98.03	8,727	LiveBench tonebench 67% token benchmark coverage
11	Claude Sonnet 4.6	24.02	91.50	7,875.5	LiveBench tonebench 67% token benchmark coverage
12	Grok 4.5	23.20	92.39	6,602.5	LiveBench tonebench 67% token benchmark coverage
13	GPT-5.6 Sol	22.38	98.30	7,342	LiveBench tonebench 67% token benchmark coverage
14	Claude Opus 4.8	22.21	96.25	13,456	LiveBench tonebench 67% token benchmark coverage
15	Mistral Large 3	22.06	79.04	2,056	tonebench 33% token benchmark coverage
16	Claude Fable 5	22.01	100.00	11,558.5	LiveBench tonebench 67% token benchmark coverage
17	GPT-5.4 Mini	20.92	85.33	25,003.5	LiveBench tonebench 67% token benchmark coverage
18	GPT-5.6 Terra	20.61	95.67	12,533	LiveBench tonebench 67% token benchmark coverage
19	Claude Sonnet 4.5	20.39	86.74	2,441	tonebench 33% token benchmark coverage
20	Command A Reasoning	20.24	70.63	2,002	tonebench 33% token benchmark coverage
21	Gemini 2.5 Flash-Lite	18.77	77.21	2,360	tonebench 33% token benchmark coverage
22	GPT-5.4	18.40	93.89	10,763.5	LiveBench tonebench 67% token benchmark coverage
23	GPT-5.6 Luna	17.55	90.25	12,547	LiveBench tonebench 67% token benchmark coverage
24	GPT-5.5	17.22	94.61	7,580	LiveBench tonebench 67% token benchmark coverage
25	GPT-5	15.39	93.69	8,777.8	Aider tonebench 67% token benchmark coverage
26	GPT-5.2	14.95	89.59	7,889	LiveBench tonebench 67% token benchmark coverage
27	GPT-5 Mini	12.49	85.41	3,925	tonebench 33% token benchmark coverage
28	GPT-5.4 Nano	12.37	82.29	25,005	LiveBench tonebench 67% token benchmark coverage
29	GPT-5.1	12.33	85.31	3,970	tonebench 33% token benchmark coverage
30	Grok 4.3	12.21	82.07	8,169	LiveBench tonebench 67% token benchmark coverage
31	Muse Spark 1.1	11.26	90.73	9,844.5	LiveBench tonebench 67% token benchmark coverage
32	Gemini 2.5 Pro	11.15	87.88	4,521	tonebench 33% token benchmark coverage
33	Gemini 2.5 Flash	10.57	81.04	4,399	tonebench 33% token benchmark coverage
34	gpt-oss-120b	10.01	47.47	3,806.6	Aider 33% token benchmark coverage
35	Inkling	8.84	89.16	5,785	tonebench 33% token benchmark coverage
36	Qwen3.7 Max	8.82	90.39	10,772.5	LiveBench tonebench 67% token benchmark coverage
37	Gemini 3.1 Pro	7.95	92.03	11,694.5	LiveBench tonebench 67% token benchmark coverage
38	Gemini 3.7 Flash	7.18	91.80	15,874.5	LiveBench tonebench 67% token benchmark coverage
39	Gemini 3.6 Flash	7.04	89.02	12,837.5	LiveBench tonebench 67% token benchmark coverage
40	Kimi-K3	6.51	96.32	14,335	LiveBench tonebench 67% token benchmark coverage
41	Grok 4.6	5.92	96.09	15,864.5	LiveBench tonebench 67% token benchmark coverage
42	GPT-5 Nano	5.32	69.19	7,465	tonebench 33% token benchmark coverage
43	Gemini 3.5 Flash	4.88	86.83	17,067	LiveBench tonebench 67% token benchmark coverage
44	Qwen3.8-Max	3.98	92.81	22,485.5	LiveBench tonebench 67% token benchmark coverage
45	DeepSeek-V4-Pro-0813	2.86	94.36	18,922	tonebench 33% token benchmark coverage
46	Claude Opus 4.5	2.34	89.41	15,203	LiveBench 33% token benchmark coverage
47	MiniMax-M3	2.08	81.18	15,530	LiveBench 33% token benchmark coverage
48	Muse Spark 1.2	2.01	94.60	18,700	LiveBench 33% token benchmark coverage
49	GLM-5.2	1.51	88.92	23,463	LiveBench 33% token benchmark coverage
50	Kimi-K2.6	1.26	85.71	27,001	LiveBench 33% token benchmark coverage
51	Qwen3.8-27B	1.25	90.15	28,740	LiveBench 33% token benchmark coverage
52	DeepSeek-V4-Pro	1.01	88.72	35,014	LiveBench 33% token benchmark coverage
53	GLM-5.3-Flash	0.97	84.66	34,707	LiveBench 33% token benchmark coverage
54	DeepSeek-V4-Flash	0.93	80.50	34,434	LiveBench 33% token benchmark coverage
55	DeepSeek-V4-Flash-Vision-Exp	0.70	92.05	52,644	LiveBench 33% token benchmark coverage
56	GLM-5.3	0.59	91.45	62,090	LiveBench 33% token benchmark coverage

Rank is global across models with published token accounting. Each token-accounted benchmark gives 100 points to its most efficient run; missing token benchmarks earn no points.

How Performance Is Aggregated

Each benchmark leader is set to 100. A model earns weighted points from every benchmark where it has a verified result; missing benchmarks earn no points, and the total produces one overall rank. Missing benchmarks are never estimated. They earn no ranking points, and coverage is shown beside every score.

How Efficiency Is Aggregated

Within each token-accounted benchmark, performance points per 1,000 output tokens are scored relative to the most efficient eligible run. Missing token-accounted benchmarks earn no points, and the total produces one overall rank. The graph uses the benchmark with the strongest configuration coverage and includes every published configuration. Highlighted points are not dominated on both benchmark performance and actual output-token spend.

How Estimated Cost Is Calculated

LiveBench quality is compared with estimated output cost using the lowest currently published USD output-token rate for each model. Score-adjusted output cost divides the estimated output cost for one published benchmark case by the model's LiveBench score as a fraction of 100. It is an estimate from published price and token artifacts, not an invoice or a task success rate.