Skip to main content
BenchLM
Advertise

Benchmark profile

Critical Physics Tasks (CritPt)

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

Data verified

Benchmark score on CritPt — July 23, 2026

BenchLM mirrors the published score view for CritPt. GPT-5.6 Sol leads the public snapshot at 32.3% , followed by GPT-5.5 Pro (30.6%) and GPT-5.6 Terra (30.0%). BenchLM does not use these results to rank models overall.

1Closed

GPT-5.6 Sol

OpenAI

gpt-5-6-sol

32.3%
Overall 81.96Context 1M
2Closed

GPT-5.5 Pro

OpenAI

gpt-5-5-pro

30.6%
Overall 63.69Context 1M
3Closed

GPT-5.6 Terra

OpenAI

gpt-5-6-terra

30.0%
Overall 72.57Context 1M
157 modelsReasoningCurrentDisplay onlyUpdated July 23, 2026

Benchmark score table (157 models)

Score
1
GPT-5.6 SolOpenAI · Closed
32.3%
2
GPT-5.5 ProOpenAI · Closed
30.6%
3
GPT-5.6 TerraOpenAI · Closed
30.0%
4
GPT-5.4 ProOpenAI · Closed
30.0%
5
Claude Fable 5Anthropic · Closed
28.6%
6
GPT-5.5OpenAI · Closed
27.1%
7
Gemini 3 Pro Deep ThinkGoogle · Closed
25.7%
8
Kimi K3Moonshot AI · Closed
23.4%
9
GPT-5.4OpenAI · Closed
23.4%
10
Claude Opus 4.8Anthropic · Closed
20.9%
11
GLM-5.2Z.AI · Open weight
20.9%
12
GPT-5.6 LunaOpenAI · Closed
20.6%
13
Gemini 3.1 ProGoogle · Closed
17.7%
14
Claude Sonnet 5Anthropic · Closed
16.9%
15
GPT-5.3 CodexOpenAI · Closed
16.9%
16
GPT-5.3-Codex-SparkOpenAI · Closed
16.9%
17
Grok 4.5xAI · Closed
15.4%
18
Muse Spark 1.1Meta · Closed
15.1%
19
Qwen3.7 MaxAlibaba · Closed
13.4%
20
Gemini 3.5 FlashGoogle · Closed
13.1%
21
DeepSeek V4 Pro (Max)DeepSeek · Open weight
12.9%
22
Claude Opus 4.6 (Adaptive)Anthropic · Closed
12.6%
23
Claude Opus 4.7 (Adaptive)Anthropic · Closed
12.0%
24
GPT-5.2OpenAI · Closed
11.6%
25
Muse SparkMeta · Closed
11.3%
26
Gemini 3.6 FlashGoogle · Closed
10.6%
27
DeepSeek V4 Pro (High)DeepSeek · Open weight
10.0%
28
GPT-5.4 miniOpenAI · Closed
10.0%
29
Kimi K2.7 CodeMoonshot AI · Open weight
10.0%
30
GPT-5.4 nanoOpenAI · Closed
9.3%
31
Qwen3.7 PlusAlibaba · Closed
9.1%
32
Gemini 3 ProGoogle · Closed
9.1%
33
GPT-5.2-CodexOpenAI · Closed
8.7%
34
Kimi K2.6Moonshot AI · Open weight
8.0%
35
Grok 4.3xAI · Closed
8.0%
36
DeepSeek V4 Flash (Max)DeepSeek · Open weight
7.1%
37
GPT-5 (high)OpenAI · Closed
5.7%
38
GPT-5.1-Codex-MaxOpenAI · Closed
5.7%
39
GPT-5.1-CodexOpenAI · Closed
5.7%
40
InklingThinking Machines Lab · Open weight
5.4%
41
Claude Opus 4.7Anthropic · Closed
5.1%
42
GPT-5.1OpenAI · Closed
4.9%
43
Hy3 PreviewTencent · Open weight
4.9%
44
Hy3Tencent · Open weight
4.9%
45
GLM-5.1Z.AI · Open weight
4.6%
46
Claude Opus 4.5 ThinkingAnthropic · Closed
4.6%
47
MiMo-V2.5-ProXiaomi · Closed
4.0%
48
MiniMax M3MiniMax · Open weight
3.7%
49
Qwen 3.6 Max (preview)Alibaba · Closed
3.7%
50
DeepSeek V4 Flash (High)DeepSeek · Open weight
3.4%
51
Kimi K2.5Moonshot AI · Open weight
3.1%
52
Nemotron 3 UltraNVIDIA · Open weight
3.1%
53
Kimi K2.5 (Reasoning)Moonshot AI · Closed
3.1%
54
Nemotron 3 Super 120B A12BNVIDIA · Open weight
3.1%
55
Qwen3.6 PlusAlibaba · Closed
2.9%
56
Grok 4.1 Fast (Reasoning)xAI · Closed
2.9%
57
Grok 4 Fast (Reasoning)xAI · Closed
2.9%
58
Claude Opus 4.6Anthropic · Closed
2.8%
59
Gemini 2.5 ProGoogle · Closed
2.6%
60
Step 3.7 FlashStepFun · Open weight
2.3%
61
Step 3.5 FlashStepFun · Open weight
2.3%
62
GLM-5Z.AI · Open weight
2.0%
63
Grok 4xAI · Closed
2.0%
64
DeepSeek V3.1 (Reasoning)DeepSeek · Open weight
2.0%
65
Qwen3.5 397BAlibaba · Open weight
1.7%
66
GLM-4.7Z.AI · Open weight
1.7%
67
Qwen3.5 397B (Reasoning)Alibaba · Open weight
1.7%
68
Gemini 3 FlashGoogle · Closed
1.4%
69
Gemma 4 31BGoogle · Open weight
1.4%
70
Gemini 2.5 FlashGoogle · Closed
1.4%
71
DeepSeek-R1DeepSeek · Open weight
1.4%
72
GPT-OSS 20BOpenAI · Open weight
1.4%
73
Qwen3.6-27BAlibaba · Open weight
1.1%
74
Claude 4 SonnetAnthropic · Closed
1.1%
75
Gemini 3.1 Flash-LiteGoogle · Closed
1.1%
76
o3OpenAI · Closed
1.1%
77
MiMo-V2-OmniXiaomi · Closed
1.1%
78
GPT-OSS 120BOpenAI · Open weight
1.1%
79
K-ExaoneLG AI Research · Closed
1.1%
80
MiniMax M2.5MiniMax · Closed
1.1%
81
Claude Sonnet 4.6Anthropic · Closed
0.9%
82
Qwen3.5-27BAlibaba · Open weight
0.9%
83
Qwen3.5-35B-A3BAlibaba · Open weight
0.9%
84
DeepSeek V3.2DeepSeek · Open weight
0.9%
85
Trinity-Large-PreviewArcee AI · Open weight
0.9%
86
Trinity-Large-ThinkingArcee AI · Open weight
0.9%
87
Nemotron 3 Nano 30BNVIDIA · Open weight
0.9%
88
Mercury 2Inception · Closed
0.8%
89
Qwen3.5-122B-A10BAlibaba · Open weight
0.6%
90
MiniMax M2.7MiniMax · Open weight
0.6%
91
GLM-5V-TurboZ.AI · Closed
0.6%
92
Gemma 4 E4BGoogle · Open weight
0.6%
93
Claude Opus 4.5Anthropic · Closed
0.3%
94
Qwen3.6-35B-A3BAlibaba · Open weight
0.3%
95
o1OpenAI · Closed
0.3%
96
Command A+Cohere · Open weight
0.3%
97
GLM-5-TurboZ.AI · Closed
0.3%
98
MiMo-V2-ProXiaomi · Closed
0.3%
99
Mistral Small 4Mistral · Open weight
0.3%
100
Sarvam 30BSarvam · Open weight
0.3%
101
Mistral Small 4 (Reasoning)Mistral · Open weight
0.3%
102
GLM-4.7-FlashZ.AI · Open weight
0.3%
103
Exaone 4.0 32BLG AI Research · Open weight
0.0%
104
Mistral Medium 3.5 128BMistral · Open weight
0.0%
105
MiMo-V2-FlashXiaomi · Open weight
0.0%
106
Nemotron 3 Nano Omni 30B A3BNVIDIA · Open weight
0.0%
107
Gemini 3.5 Flash-LiteGoogle · Closed
0.0%
108
LFM2.5-8B-A1BLiquidAI · Open weight
0.0%
109
Kimi K2Moonshot AI · Closed
0.0%
110
Gemma 4 26B A4BGoogle · Open weight
0.0%
111
GPT-4.1 nanoOpenAI · Closed
0.0%
112
GPT-4.1OpenAI · Closed
0.0%
113
GPT-4.1 miniOpenAI · Closed
0.0%
114
Gemma 4 12BGoogle · Open weight
0.0%
115
GLM-4.6Z.AI · Open weight
0.0%
116
DeepSeek V3DeepSeek · Open weight
0.0%
117
GPT-4oOpenAI · Closed
0.0%
118
Llama 4 ScoutMeta · Open weight
0.0%
119
Llama 4 MaverickMeta · Open weight
0.0%
120
Ling 2.6 FlashInclusionAI · Open weight
0.0%
121
Grok 4.1 FastxAI · Closed
0.0%
122
DeepSeek V3.1DeepSeek · Open weight
0.0%
123
Mistral Large 3Mistral · Closed
0.0%
124
GLM-4.5-AirZ.AI · Closed
0.0%
125
Gemma 3 27BGoogle · Open weight
0.0%
126
GPT-5 (medium)OpenAI · Closed
0.0%
127
Mistral Large 2Mistral · Closed
0.0%
128
Llama 3.1 405BMeta · Open weight
0.0%
129
Phi-4Microsoft · Open weight
0.0%
130
Grok Code Fast 1xAI · Closed
0.0%
131
Nemotron Ultra 253BNVIDIA · Open weight
0.0%
132
Claude 3 HaikuAnthropic · Closed
0.0%
133
Claude 4.1 Opus ThinkingAnthropic · Closed
0.0%
134
Nova ProAmazon · Closed
0.0%
135
LFM2.5-VL-1.6B-ExtractLiquidAI · Open weight
0.0%
136
Qwen3 MaxAlibaba · Closed
0.0%
137
Gemma 4 E2BGoogle · Open weight
0.0%
138
Sarvam 105BSarvam · Open weight
0.0%
139
Mistral Medium 3Mistral · Closed
0.0%
140
Granite-4.0-1BIBM · Open weight
0.0%
141
Granite-4.0-350MIBM · Open weight
0.0%
142
Granite-4.0-H-1BIBM · Open weight
0.0%
143
Granite-4.0-H-350MIBM · Open weight
0.0%
144
Exaone 4.0 1.2BLG AI Research · Open weight
0.0%
145
Solar Pro 2Upstage · Closed
0.0%
146
GPT-5 miniOpenAI · Closed
0.0%
147
Ministral 3 14B (Reasoning)Mistral · Open weight
0.0%
148
Ministral 3 14BMistral · Open weight
0.0%
149
GPT-5 nanoOpenAI · Closed
0.0%
150
MiniMax M1 80kMiniMax · Closed
0.0%
151
LFM2-24B-A2BLiquidAI · Closed
0.0%
152
Ministral 3 8B (Reasoning)Mistral · Open weight
0.0%
153
LFM2.5-1.2B-ThinkingLiquidAI · Closed
0.0%
154
Ministral 3 8BMistral · Open weight
0.0%
155
Ministral 3 3B (Reasoning)Mistral · Open weight
0.0%
156
LFM2.5-1.2B-InstructLiquidAI · Closed
0.0%
157
Ministral 3 3BMistral · Open weight
0.0%

The published CritPt snapshot places GPT-5.6 Sol first at 32.3%. The third row is 2.3 points behind. The broader top-10 range is 11.4 points, so the table still separates the published systems.

157 models have been evaluated on CritPt. The benchmark falls in the Reasoning category. This category carries a 17% weight in BenchLM.ai's overall scoring system. CritPt is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About CritPt

Year

2026

Tasks

Research-level physics questions

Format

Accuracy

Difficulty

Research-level physics reasoning

BenchLM stores CritPt as a display-only research-physics reasoning row from Artificial Analysis' independently evaluated leaderboard.

CritPt Benchmark Leaderboard

BenchLM freshness & provenance

Version

CritPt 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does CritPt measure?

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

Which model scores highest on CritPt?

GPT-5.6 Sol by OpenAI currently leads with a score of 32.3% on CritPt.

How many models are evaluated on CritPt?

157 AI models have been evaluated on CritPt on BenchLM.

Compare Top Models on CritPt

GPT-5.6 Sol vs GPT-5.5 ProGPT-5.5 Pro vs GPT-5.6 TerraGPT-5.6 Terra vs GPT-5.4 ProGPT-5.4 Pro vs Claude Fable 5
Last updated: July 23, 2026 · BenchLM version CritPt 2026

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.