CritPt leaderboard

A display-only Artificial Analysis metric for research-level physics reasoning.

Models ranked by their published Critical Physics Tasks value

Rank	Model	Provider	Accuracy
1	GPT-5.6 Sol	OpenAI	32.3
2	GPT-6 Astra	OpenAI	31.7
3	GPT-5.5 Pro	OpenAI	30.6
4	GPT-5.4 Pro	OpenAI	30.0
4	GPT-5.6 Terra	OpenAI	30.0
6	Claude Fable 5.1	Anthropic	29.7
7	Claude Opus 5	Anthropic	29.1
8	Claude Fable 5	Anthropic	28.6
9	GPT-5.5	OpenAI	27.1
10	Gemini 3 Pro Deep Think	Google	25.7
11	Muse Spark 1.3	Meta	24.9
12	GPT-5.4	OpenAI	23.4
12	Kimi K3	Moonshot AI	23.4
14	Claude Opus 4.8	Anthropic	20.9
14	GLM-5.2	Z.AI	20.9
16	GPT-5.6 Luna	OpenAI	20.6
17	GLM-5.3	Z.AI	19.1
18	Gemini 3.8 Flash	Google	18.3
19	DeepSeek V4 Pro 0813	DeepSeek	18.0
20	Gemini 3.1 Pro	Google	17.7
20	Muse Spark 1.2	Meta	17.7
20	Qwen3.8 Max Preview	Alibaba	17.7
23	Grok 4.6	xAI	17.1
24	Claude Sonnet 5	Anthropic	16.9
24	GPT-5.3 Codex	OpenAI	16.9
24	Hy4 preview	Tencent	16.9
27	DeepSeek V4 Flash 0731	DeepSeek	16.6
28	Grok 4.5	xAI	15.4
29	Muse Spark 1.1	Meta	15.1
30	DeepSeek V4.1 Flash	DeepSeek	14.3
30	Gemini 3.7 Flash	Google	14.3
32	Qwen3.7 Max	Alibaba	13.4
33	Gemini 3.5 Flash	Google	13.1
34	Claude Opus 4.6 (Adaptive)	Anthropic	12.6
35	Claude Opus 4.7 (Adaptive)	Anthropic	12.0
36	GPT-5.2	OpenAI	11.6
37	Muse Spark	Meta	11.3
38	Qwen3.8-Flash-Next	Alibaba	11.1
39	Gemini 3.6 Flash	Google	10.6
40	GPT-5.4 mini	OpenAI	10.0
40	Kimi K2.7 Code	Moonshot AI	10.0
42	MCQuasar 438B	Multiverse Computing	9.4
43	GPT-5.4 nano	OpenAI	9.3
44	Gemini 3 Pro	Google	9.1
44	Qwen3.7 Plus	Alibaba	9.1
46	STA.X K2	SK Telecom	8.9
47	GPT-5.2-Codex	OpenAI	8.7
48	TMInkling-Small	Thinking Machines Lab	8.3
49	Grok 4.3	xAI	8.0
49	Kimi K2.6	Moonshot AI	8.0
51	GPT-5 (high)	OpenAI	5.7
51	GPT-5.1-Codex	OpenAI	5.7
51	GPT-5.1-Codex-Max	OpenAI	5.7
54	TMInkling	Thinking Machines Lab	5.4
54	Qwen3.8-27B	Alibaba	5.4
54	Solar Pro 4	Upstage	5.4
57	Claude Opus 4.7	Anthropic	5.1
58	GPT-5.1	OpenAI	4.9
58	Hy3	Tencent	4.9
58	Hy3 Preview	Tencent	4.9
61	APApodex 1.1	Apodex	4.6
61	APApodex 1.1 Mini	Apodex	4.6
61	Claude Opus 4.5 Thinking	Anthropic	4.6
61	GLM-5.1	Z.AI	4.6
65	MiMo-V2.5-Pro	Xiaomi	4.0
66	MiniMax M3	MiniMax	3.7
66	Qwen 3.6 Max (preview)	Alibaba	3.7
68	Kimi K2.5	Moonshot AI	3.1
68	Kimi K2.5 (Reasoning)	Moonshot AI	3.1
68	Nemotron 3 Super 100B	NVIDIA	3.1
68	Nemotron 3 Ultra	NVIDIA	3.1
72	Grok 4 Fast (Reasoning)	xAI	2.9
72	Grok 4.1 Fast (Reasoning)	xAI	2.9
72	Qwen3.6 Plus	Alibaba	2.9
75	Claude Opus 4.6	Anthropic	2.8
76	Gemini 2.5 Pro	Google	2.6
76	Muse Glimmer 30B	Meta	2.6
78	Step 3.7 Flash	StepFun	2.3
79	DeepSeek V3.1 (Reasoning)	DeepSeek	2.0
79	GLM-5	Z.AI	2.0
79	Grok 4	xAI	2.0
79	Ling 3.0 Flash VL	InclusionAI	2.0
83	GLM-4.7	Z.AI	1.7
83	Ling 3.0 Flash	InclusionAI	1.7
83	Ling 3.0 Flash FP8	InclusionAI	1.7
86	DeepSeek-R1	DeepSeek	1.4
86	Gemini 2.5 Flash	Google	1.4
86	Gemini 3 Flash	Google	1.4
86	Gemma 4 31B	Google	1.4
86	GPT-OSS 20B	OpenAI	1.4
91	Claude 4 Sonnet	Anthropic	1.1
91	GPT-OSS 120B	OpenAI	1.1
91	K-Exaone	LG AI Research	1.1
91	MiMo-V2-Omni	Xiaomi	1.1
91	o3	OpenAI	1.1
91	Qwen3.6-27B	Alibaba	1.1
97	Claude Sonnet 4.6	Anthropic	0.9
97	DeepSeek V3.2	DeepSeek	0.9
97	K-EXAONE 2.0	LG AI Research	0.9
97	Nemotron 3 Nano 30B	NVIDIA	0.9
97	Qwen3.5 397B	Alibaba	0.9
97	Qwen3.5 397B (Reasoning)	Alibaba	0.9
97	Qwen3.5-27B	Alibaba	0.9
97	Qwen3.5-35B-A3B	Alibaba	0.9
97	Trinity-Large-Preview	Arcee AI	0.9
97	Trinity-Large-Thinking	Arcee AI	0.9
107	Gemma 4 E4B	Google	0.6
107	GLM-5V-Turbo	Z.AI	0.6
107	MiniMax M2.7	MiniMax	0.6
107	Qwen3.5-122B-A10B	Alibaba	0.6
111	Claude Opus 4.5	Anthropic	0.3
111	Command A+	Cohere	0.3
111	GLM-5-Turbo	Z.AI	0.3
111	Granite 4.2 30B	IBM	0.3
111	Granite 4.2 8B	IBM	0.3
111	MiMo-V2-Pro	Xiaomi	0.3
111	OPMiniCPM5-2B	OpenBMB	0.3
111	Mistral Small 4	Mistral	0.3
111	Mistral Small 4 (Reasoning)	Mistral	0.3
111	North Mini Code	Cohere	0.3
111	o1	OpenAI	0.3
111	Qwen3.6-35B-A3B	Alibaba	0.3
111	SASarvam 30B	Sarvam	0.3
124	CECeleris-1	Celeris	0.0
124	Claude 3 Haiku	Anthropic	0.0
124	Claude 4.1 Opus Thinking	Anthropic	0.0
124	DeepSeek V3	DeepSeek	0.0
124	DeepSeek V3.1	DeepSeek	0.0
124	Exaone 4.0 1.2B	LG AI Research	0.0
124	Exaone 4.0 32B	LG AI Research	0.0
124	Gemini 3.5 Flash-Lite	Google	0.0
124	Gemma 3 27B	Google	0.0
124	Gemma 4 12B	Google	0.0
124	Gemma 4 26B A4B	Google	0.0
124	Gemma 4 E2B	Google	0.0
124	GLM-4.5-Air	Z.AI	0.0
124	GLM-4.6	Z.AI	0.0
124	GPT-4.1	OpenAI	0.0
124	GPT-4.1 mini	OpenAI	0.0
124	GPT-4.1 nano	OpenAI	0.0
124	GPT-4o	OpenAI	0.0
124	GPT-5 (medium)	OpenAI	0.0
124	Granite 4.2 3B	IBM	0.0
124	Granite-4.0-1B	IBM	0.0
124	Granite-4.0-350M	IBM	0.0
124	Granite-4.0-H-1B	IBM	0.0
124	Granite-4.0-H-350M	IBM	0.0
124	Grok 4.1 Fast	xAI	0.0
124	Grok Code Fast 1	xAI	0.0
124	Kimi K2	Moonshot AI	0.0
124	LFM2.5-2.6B	LiquidAI	0.0
124	LFM2.5-8B-A1B	LiquidAI	0.0
124	LFM2.5-VL-1.6B-Extract	LiquidAI	0.0
124	Ling 2.6 Flash	InclusionAI	0.0
124	Ling 3.0 Tiny	InclusionAI	0.0
124	Llama 3.1 405B	Meta	0.0
124	Llama 4 Maverick	Meta	0.0
124	Llama 4 Scout	Meta	0.0
124	MiMo-V2-Flash	Xiaomi	0.0
124	Mistral Large 2	Mistral	0.0
124	Mistral Large 3	Mistral	0.0
124	Mistral Medium 3	Mistral	0.0
124	Mistral Medium 3.5 128B	Mistral	0.0
124	Nemotron 3 Nano Omni 30B A3B	NVIDIA	0.0
124	Nemotron 3.5 Lightning 30B A3B NVFP4	NVIDIA	0.0
124	Nemotron Ultra 253B	NVIDIA	0.0
124	Nova Pro	Amazon	0.0
124	Phi-4	Microsoft	0.0
124	Qwen3 Max	Alibaba	0.0
124	Qwen3-Omni-30B-A3B-Instruct	Alibaba	0.0
124	Qwen3-Omni-30B-A3B-Thinking	Alibaba	0.0
124	SASarvam 105B	Sarvam	0.0
124	Solar Pro 2	Upstage	0.0
124	Solar Pro 3	Upstage	0.0
124	FAUltravox v0.6 Llama 3.3 70B	Fixie AI	0.0

Evidence key: ObservedLast good

Rows are ordered by the value CritPt Benchmark Leaderboard published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.