Benchmarks
159 benchmarks em 11 categorias. Clique para ver a classificação completa.
Programação
Aider — Code Editing
Programação1
o1
84.2
2
Claude 3.5 Sonnet
84.2
3o1-preview
79.7
4GPT-4o (2024-05-13)
72.9
5GPT-4o (2024-11-20)
71.4
29 modelos testados
Aider polyglot
Programação1GPT-5 Chat
88.0
2GPT-5
88.0
3o3 Pro
84.9
4
Gemini 2.5 Pro
83.1
5Gemini 2.5 Pro Preview 06-05
83.1
54 modelos testados
CadEval
Programação1o3
74.0
2Gemini 2.5 Pro
64.0
3o4 Mini
62.0
4o1
56.0
5Claude 3.7 Sonnet
54.0
17 modelos testados
Cybench
Programação1Claude Opus 4.6
93.0
2Claude Opus 4.5
82.0
3Claude Sonnet 4.5
60.0
4
Grok 4
43.0
5Claude Opus 4.1
42.0
21 modelos testados
Deepswe
Programação1GPT-6 Astra
74.1
2Gemini 3.8 Flash
73.8
3Claude Opus 5
73.7
4GPT-5.6 Sol
72.7
5Claude Fable 5
69.9
23 modelos testados
Frontiercode
Programação1Claude Opus 5.5
54.6
2Claude Fable 5
53.5
3Claude Opus 5
53.4
4GPT-6 Astra
53.3
5Claude Sonnet 5.5
52.1
40 modelos testados
Frontierswe
Programação1GPT-6 Astra
65.5
2Claude Opus 5.5
62.3
3Claude Sonnet 5.5
61.9
4Claude Fable 5.1
56.3
5Gemini 4 Argon
55.0
17 modelos testados
GSO-Bench
Programação1Claude Fable 5.1
88.2
2GPT-6 Astra
79.4
3Claude Fable 5
78.4
4GPT-5.6 Sol
76.5
5Claude Opus 4.8
47.1
27 modelos testados
LiveBench — Agentic Coding
Programação1
GLM 5.2
73.3
2
Kimi K2.7 Code
70.0
3
MiniMax M3
60.0
4
DeepSeek V4 Pro
56.7
5GPT-5.1-Codex-Max
56.7
40 modelos testados
LiveBench — Coding
Programação1GPT-5.2-Codex
83.6
2GPT-5.1-Codex-Max
81.4
3GLM 5.2
79.7
4
Qwen3.6 Plus
78.2
5GPT-5 Mini
76.1
40 modelos testados
Mirrorcode
Programação1Claude Opus 5.5
77.4
2Claude Fable 5.1
73.3
3Claude Fable 5
63.9
4GPT-6 Astra
46.7
5Claude Opus 4.7
31.1
8 modelos testados
OpenCompass — LiveCodeBenchV6
Programação1GLM 5
86.2
2
Step 3.5 Flash
83.9
3GLM 4.7
83.8
4Qwen3.5 397B A17B
83.0
5DeepSeek V3.2 Speciale
80.9
34 modelos testados
SWE-Bench verified
Programação1Claude Mythos Preview
93.9
2Claude Opus 4.7
83.5
3Gemini 3.5 Flash
79.3
4Claude Opus 4.6
78.7
5GLM 5.2
78.7
32 modelos testados
SWE-Bench Verified (Bash Only)
Programação1Claude Opus 4.5
74.4
2GPT-5.2
71.8
3Claude Sonnet 4.5
70.6
4Claude Opus 4
67.6
5GPT-5.1
66.0
19 modelos testados
Terminal Bench
Programação1GPT-5.5
82.7
2Claude Mythos Preview
82.0
3GPT-5.4
81.8
4Gemini 3.1 Pro Preview
80.2
5Claude Opus 4.7
80.2
36 modelos testados
WeirdML
Programação1GPT-6 Astra
93.6
2Claude Fable 5.1
92.9
3Claude Fable 5
91.9
4Claude Opus 5
91.8
5GPT-5.6 Sol
89.4
116 modelos testados
Conhecimento
ANLI
Conhecimento1GPT-3.5 Turbo (older v0613)
37.1
2
phi-3-small 7.4B
37.1
3
Llama 3-70B
36.0
4Llama 3 8B Instruct
36.0
5phi-3-medium 14B
33.7
11 modelos testados
ARC AI2
Conhecimento1Llama 3.1 405B
93.7
2DeepSeek V3
93.7
3Qwen2.5 72B Instruct
92.7
4DeepSeek-V2 (MoE-236B, May 2024)
89.6
5phi-3-medium 14B
88.8
52 modelos testados
AudioMultiChallenge — Audio Output
Conhecimento0 modelos testados
AudioMultiChallenge — Text Output
Conhecimento1Gemini 2.5 Pro
46.9
2Gemini 2.5 Flash
40.0
3
Voxtral Small 24B 2507
26.3
3 modelos testados
Balrog
Conhecimento1GPT-6 Astra
68.3
2Claude Opus 5
63.4
3GPT-5.6 Sol
60.0
4Gemini 3 Pro
58.1
5Gemini 3.1 Pro Preview
57.0
36 modelos testados
Chess Puzzles
Conhecimento1GPT-6 Astra
72.6
2GPT-6.1 Sol
66.3
3GPT-5.6 Sol
66.3
4Claude Opus 5.5
65.3
5GPT-5.4 Pro
58.6
121 modelos testados
CMMLU
Conhecimento1Qwen2-72B
89.7
2Qwen2.5 72B Instruct
85.7
3GPT-4 Turbo
71.0
4Llama 3.1 70B Instruct
64.4
5Qwen-14B
58.7
8 modelos testados
DeepResearch Bench
Conhecimento1Claude Opus 4.6
55.3
2GPT-5
55.1
3Claude Sonnet 4.6
54.9
4Claude Opus 4.5
54.8
5Claude Sonnet 4.5
52.6
26 modelos testados
Fiction.LiveBench
Conhecimento1o3 Pro
97.2
2GPT-5
97.2
3Grok 4
94.4
4Grok 4 Fast
94.4
5Gemini 2.5 Pro
91.7
49 modelos testados
GeoBench
Conhecimento1Gemini 3 Flash Preview
88.0
2Gemini 3 Pro
84.0
3Gemini 2.5 Pro
81.0
4GPT-5
81.0
5o1
80.0
27 modelos testados
GPQA
Conhecimento1Meta Llama 3 8B
19.7
2
HA
Qwen2.5 72B Instruct Abliterated19.4
3Qwen2-72B
19.2
4DeepSeek R1 Distill Qwen 14B
18.3
5WizardLM-2 8x22B
17.6
78 modelos testados
GPQA diamond
Conhecimento1Gemini 4 Argon
95.2
2Claude Mythos Preview
94.5
3GPT-6 Astra
94.4
4GPT-5.5 Pro
94.2
5Claude Sonnet 5.5
94.1
179 modelos testados
HellaSwag
Conhecimento1GPT-4 Turbo
93.7
2Llama 3.1 405B
85.6
3
Falcon-180B
85.3
4DeepSeek V3
85.2
5DeepSeek-V2 (MoE-236B, May 2024)
82.8
42 modelos testados
HELM — GPQA
Conhecimento1Gemini 3 Pro
80.3
2GPT-5 Chat
79.1
3GPT-5
79.1
4GPT-5 Mini
75.6
5o3
75.3
40 modelos testados
HELM — MMLU-Pro
Conhecimento1Gemini 3 Pro
90.3
2GPT-5 Chat
86.3
3Gemini 2.5 Pro
86.3
4GPT-5
86.3
5o3
85.9
40 modelos testados
HLE
Conhecimento1Claude Mythos Preview
56.8
2GPT-6 Astra
52.5
3Claude Fable 5.1
43.8
4Gemini 3.1 Pro Preview
43.7
5Claude Sonnet 5.5
43.4
39 modelos testados
Humanity's Last Exam
Conhecimento0 modelos testados
Humanity's Last Exam (Text Only)
Conhecimento0 modelos testados
LAMBADA
Conhecimento1Falcon-180B
79.8
2Llama 2-13B
76.5
3LLaMA-13B
75.2
4
U
Baichuan 2-7B73.3
5
U
internlm-20b71.8
9 modelos testados
Lech Mazur Writing
Conhecimento1Qwen3 Max
87.1
2o3 Pro
86.3
3GPT-5
86.0
4Kimi K2 0711
85.6
5Claude Opus 4.1
85.4
46 modelos testados
LiveBench — Overall
Conhecimento1GLM 5.2
76.2
2GPT-5.2-Codex
74.3
3Qwen3.7 Max
74.3
4DeepSeek V4 Pro
73.6
5GPT-5.1-Codex-Max
72.0
40 modelos testados
MMLU
Conhecimento1GPT-4o (2024-11-20)
84.1
2DeepSeek V3
82.9
3Claude 3.5 Sonnet
82.0
4GPT-4 (older v0314)
81.9
5Llama 3.3 70B Instruct
81.7
83 modelos testados
MMLU-PRO
Conhecimento1Qwen2-72B
52.6
2Qwen2.5 32B Instruct
51.9
3Qwen2.5 72B Instruct
51.4
4
HA
Qwen2.5 72B Instruct Abliterated50.4
5Phi 4
48.3
78 modelos testados
OpenBookQA
Conhecimento1phi-3-small 7.4B
84.0
2phi-3-mini 3.8B
84.0
3phi-3-medium 14B
83.2
4GPT-3.5 Turbo (older v0613)
81.3
5Mixtral 8x7B Instruct
81.1
27 modelos testados
OpenCompass — GPQA-Diamond
Conhecimento1Qwen3.5 397B A17B
88.4
2Kimi K2.5
88.1
3GLM 4.7
86.9
4DeepSeek V3.2 Speciale
86.7
5GLM 5
85.3
34 modelos testados
OpenCompass — HLE
Conhecimento1DeepSeek V3.2 Speciale
28.6
2Kimi K2.5
28.6
3GLM 5
28.1
4Qwen3.5 397B A17B
27.5
5GLM 4.7
25.4
34 modelos testados
OpenCompass — MMLU-Pro
Conhecimento1Qwen3.5 397B A17B
87.6
2Kimi K2.5
86.2
3DeepSeek V3.2
85.8
4Gemini 2.5 Pro
85.8
5DeepSeek V3.2 Speciale
85.5
34 modelos testados
PIQA
Conhecimento1GPT-4o-mini (2024-07-18)
77.4
2GPT-4o-mini
77.4
3Gemini 1.5 Flash (May 2024)
75.0
4Llama 3.1 405B
71.8
5Falcon-180B
69.8
33 modelos testados
PostTrainBench
Conhecimento1Claude Fable 5
41.8
2GPT-5.6 Sol
36.2
3Claude Opus 5
35.0
4Claude Opus 4.8
33.8
5Kimi K3
32.0
23 modelos testados
Professional Reasoning — Finance
Conhecimento1Claude Opus 4.6 (Fast)
53.3
2Claude Opus 4.6
53.3
3GPT-5
51.3
4GPT-5 Pro
51.1
5o3 Pro
49.1
6 modelos testados
Professional Reasoning — Legal
Conhecimento1Claude Opus 4.6 (Fast)
52.3
2Claude Opus 4.6
52.3
3GPT-5 Pro
49.9
4o3 Pro
49.7
5GPT-5
49.0
6 modelos testados
ScienceQA
Conhecimento1GPT-4o (2024-05-13)
84.7
2GPT-4o (2024-11-20)
84.7
3Claude 3 Haiku
62.7
4Llama 2-13B
41.0
5LLaMA-13B
24.4
5 modelos testados
SimpleQA Verified
Conhecimento1Gemini 4 Argon
76.0
2GPT-6 Astra
75.6
3GPT-6.1 Sol
73.9
4Gemini 3.1 Pro Preview
73.5
5Gemini 3 Pro
72.9
87 modelos testados
SpireClimb
Conhecimento1GPT-6 Astra
25.7
2GPT-6.1 Sol
24.7
3GPT-6 Sol
7.5
4Claude Opus 5.5
3.3
5Claude Fable 5.1
2.0
7 modelos testados
TriviaQA
Conhecimento1Claude 2
87.5
2
U
PaLM 2-L86.1
3GPT-3.5 Turbo (older v0613)
85.8
4GPT-4 Turbo
84.8
5DeepSeek V3
82.9
25 modelos testados
VisualToolBench (VTB)
Conhecimento1Gemini 3.1 Pro Preview
29.0
2Claude Opus 4.6 (Fast)
27.5
3Claude Opus 4.6
27.5
3 modelos testados
VPCT
Conhecimento1Gemini 3 Pro
86.5
2GPT-5.2
76.0
3Gemini 3 Flash Preview
58.9
4GPT-5
49.0
5GPT-5.1
38.0
23 modelos testados
Winogrande
Conhecimento1Llama 3.1 405B
78.4
2Claude 3 Opus
77.0
3GPT-4 (older v0314)
75.0
4GPT-4 Turbo
75.0
5Falcon-180B
74.2
53 modelos testados
Agêntico
APEX-Agents
Agêntico1Gemini 4 Argon
82.4
2Claude Sonnet 5.5
75.5
3Claude Opus 5.5
73.5
4Claude Fable 5.1
68.6
5Gemini 3.7 Flash
67.8
69 modelos testados
OSWorld
Agêntico1Claude Mythos Preview
79.6
2GPT-5.5
78.7
3Claude Sonnet 4.6
72.1
4Claude Opus 4.5
66.3
5Kimi K2.5
63.3
10 modelos testados
SWE Atlas — Codebase QnA
Agêntico1Claude Opus 4.6 (Fast)
33.3
2Claude Opus 4.6
33.3
3GPT-5.3-Codex
32.6
4Claude Sonnet 4.6
31.2
4 modelos testados
SWE Atlas — Test Writing
Agêntico1Claude Opus 4.6 (Fast)
36.7
2Claude Opus 4.6
36.7
3Claude Sonnet 4.6
31.8
3 modelos testados
SWE-Bench Pro (Private)
Agêntico1Claude Opus 4.5
23.4
2Gemini 2.5 Pro Preview 06-05
10.1
2 modelos testados
The Agent Company
Agêntico1DeepSeek V3.2 Exp
42.9
2Gemini 2.5 Flash
41.1
3Claude Sonnet 4
33.1
4Claude 3.7 Sonnet
30.9
5Gemini 2.5 Pro
30.3
15 modelos testados
Raciocínio
ARC-AGI
Raciocínio1Claude Fable 5
98.5
2GPT-6 Astra
98.5
3Claude Opus 5.5
98.5
4GPT-6.1 Sol
98.5
5Gemini 3.8 Flash
98.5
80 modelos testados
ARC-AGI-2
Raciocínio1GPT-6 Astra
95.0
2GPT-6.1 Sol
94.2
3Claude Opus 5.5
93.3
4GPT-5.6 Sol
92.5
5Claude Opus 5
90.4
83 modelos testados
BBH
Raciocínio1Gemini 1.5 Pro (May 2024)
85.6
2DeepSeek V3
83.3
3Gemini 1.5 Pro (Feb 2024)
78.7
4Llama 3.1 405B
77.2
5phi-3-medium 14B
75.2
28 modelos testados
HELM — WildBench
Raciocínio1GPT-5.1
86.3
2Kimi K2 0711
86.2
3o3
86.1
4Gemini 3 Pro
85.9
5GPT-5 Chat
85.7
40 modelos testados
LiveBench — Data Analysis
Raciocínio1GPT-5.2-Codex
78.2
2MiniMax M3
76.2
3DeepSeek V4 Pro
74.5
4GLM 5.2
73.7
5Qwen3.7 Max
71.8
40 modelos testados
LiveBench — Reasoning
Raciocínio1GPT-5.1-Codex-Max
84.6
2Qwen3.7 Max
83.3
3Kimi K2.7 Code
82.8
4DeepSeek V4 Pro
82.7
5GPT-5.1-Codex
82.0
40 modelos testados
MUSR
Raciocínio1DeepSeek R1 Distill Qwen 14B
28.7
2
Hermes 3 70B Instruct
23.4
3Llama 3 8B Instruct
19.9
4Qwen2-72B
19.7
5
U
Stable Beluga 218.6
78 modelos testados
SimpleBench
Raciocínio1Claude Fable 5
78.3
2Claude Opus 5
76.7
3Gemini 3.1 Pro Preview
75.5
4Gemini 3.5 Flash
72.0
5Gemini 3 Pro
71.7
78 modelos testados
Velocidade
Artificial Analysis — Agentic Index
Velocidade1GPT-5.4
69.4
2Claude Opus 4.6 (Fast)
67.6
3GLM 5 Turbo
63.1
4
MiMo-V2-Pro
62.8
5GPT-5.3-Codex
62.2
88 modelos testados
Artificial Analysis — Coding Index
Velocidade1Claude Fable 5
76.5
2Claude Opus 4.8 (Fast)
74.3
3Claude Opus 4.7 (Fast)
73.6
4Gemini 3.5 Flash
70.1
5Gemini 3.1 Pro Preview
68.8
89 modelos testados
Artificial Analysis — Quality Index
Velocidade1Claude Fable 5
59.9
2Claude Opus 5.5
57.6
3GPT-5.4
57.2
4Claude Sonnet 5.5
56.0
5Claude Opus 4.8 (Fast)
55.7
135 modelos testados
Artificial Analysis · CritPt
Velocidade1GPT-6 Astra
31.7
2Claude Opus 5.5
31.7
3GPT-6.1 Sol
31.7
4Claude Sonnet 5.5
31.4
5GPT-5.6 Terra
30.0
69 modelos testados
Artificial Analysis · GDPval
Velocidade1Claude Opus 5.5
68.3
2Claude Sonnet 5.5
66.9
3Claude Fable 5.1
62.7
4Grok 4.7
60.6
5MiMo-V2.6-Pro
59.2
60 modelos testados
Artificial Analysis · GPQA Diamond
Velocidade1GPT-6 Astra
96.1
2Gemini 3.8 Flash
95.3
3Gemini 3.1 Pro Preview
94.1
4Claude Fable 5.1
93.7
5Kimi K3
93.5
55 modelos testados
1Claude Opus 5.5
61.4
2Claude Fable 5.1
59.1
3Gemini 4 Argon
57.1
4Claude Sonnet 5.5
55.0
5GPT-6 Astra
54.7
70 modelos testados
Artificial Analysis · IFBench
Velocidade1MiniMax M3
82.9
2
Nemotron 3 Ultra
81.4
3Qwen3.7 Plus
78.0
4Gemini 3.1 Pro Preview
77.1
5Gemma 4 31B
75.6
37 modelos testados
1Kimi K3
88.7
2Step 5 Preview
88.3
3MiMo-V2.6-Pro
86.3
4Claude Fable 5.1
85.3
5Claude Opus 5.5
84.7
69 modelos testados
Artificial Analysis · MMMU Pro
Velocidade1Claude Opus 5.5
87.7
2GPT-6 Astra
86.9
3GPT-6.1 Sol
86.0
4Gemini 3.8 Flash
85.6
5Qwen3.8 Max (0902)
82.8
35 modelos testados
Artificial Analysis · SciCode
Velocidade1Claude Opus 5.5
66.9
2Claude Fable 5.1
63.1
3Gemini 4 Argon
61.8
4Claude Sonnet 5.5
61.0
5MiMo-V2.6-Pro
60.9
55 modelos testados
Artificial Analysis · tau2-Bench Telecom
Velocidade1Gemini 3.1 Pro Preview
95.6
2Qwen3.6 35B A3B
95.3
3Mistral Medium 3.5
94.2
4Qwen3.7 Plus
93.0
5Qwen3.5 4B
92.1
37 modelos testados
1GPT-5.6 Terra
57.6
2Gemini 3.1 Pro Preview
53.8
3GPT-5.3-Codex
53.0
4Qwen3.7 Plus
47.0
5Kimi K2.7 Code
44.7
37 modelos testados
Geral
BBH (HuggingFace)
Geral1Qwen2.5 72B Instruct
61.9
2
HA
Qwen2.5 72B Instruct Abliterated60.5
3Llama 3.3 70B Instruct
56.6
4Qwen2.5 32B Instruct
56.5
5Llama 3.1 70B Instruct
55.9
78 modelos testados
Cl Bench
Geral1GPT-5.4
27.9
2GPT-5.1
23.7
3Grok 4.20
22.2
4Claude Opus 4.5
21.1
5Gemini 3.1 Pro Preview
20.8
19 modelos testados
Cl Bench Life
Geral1GPT-5.4
21.7
2GPT-5.1
17.3
3Claude Opus 4.6
17.0
4Gemini 3.1 Pro Preview
16.9
5DeepSeek V4 Pro
13.5
12 modelos testados
Dtbench
Geral1Claude Opus 5.5
98.2
2Claude Fable 5
97.3
3Claude Opus 5
96.5
4Claude Fable 5.1
96.0
5GPT-6 Astra
95.5
135 modelos testados
Ebr Bench
Geral1GPT-6 Astra
76.2
2Claude Opus 5.5
71.4
3Claude Fable 5.1
57.1
4GPT-6.1 Sol
54.3
5GPT-6 Sol
53.3
24 modelos testados
Exploitbench
Geral1Claude Opus 4.7
26.5
2Gemini 3.1 Pro Preview
26.1
3Claude Sonnet 4.6
23.6
4Kimi K2.6
18.4
5GLM 5.1
18.1
7 modelos testados
Furniture Assembly
Geral1Claude Opus 5.5
76.2
2GPT-6 Astra
71.4
3GPT-6.1 Sol
71.4
4Claude Sonnet 5.5
64.3
5Gemini 4 Argon
60.7
32 modelos testados
Gdpval
Geral1GPT-5.2
49.7
2Claude Opus 4.5
45.5
3Claude Opus 4.1
43.6
4Claude Sonnet 4.5
42.5
5Gemini 3 Pro
40.3
11 modelos testados
Lmca
Geral1Claude Opus 5.5
80.3
2Claude Fable 5.1
77.0
3Claude Opus 5
75.8
4GPT-6 Astra
75.7
5Claude Fable 5
71.9
113 modelos testados
Metr Time Horizons
Geral1Claude Opus 4.6
78.9
2Gemini 3.1 Pro Preview
77.0
3GPT-5.2
75.3
4Claude Opus 4.5
75.0
5GPT-5.3-Codex
74.5
29 modelos testados
Mystery Game Puzzles
Geral1GPT-6 Astra
82.4
2GPT-6.1 Sol
78.0
3Claude Opus 5.5
68.0
4Claude Sonnet 5.5
61.5
5Gemini 4 Argon
60.3
73 modelos testados
Osworld 2 0
Geral1Claude Opus 5
31.4
2GPT-5.6 Sol
27.3
3Claude Opus 4.8
20.6
4Claude Opus 4.7
18.2
5Claude Sonnet 4.6
9.3
8 modelos testados
Proofbench
Geral1Claude Fable 5.1
100.0
2Claude Opus 5.5
100.0
3Claude Sonnet 5.5
100.0
4GPT-6 Astra
99.0
5Claude Opus 5
99.0
63 modelos testados
Remote Labor Index
Geral1Claude Opus 5.5
21.3
2GPT-6 Astra
20.8
3Claude Fable 5.1
17.9
4Claude Fable 5
16.1
5Claude Opus 4.8
8.3
14 modelos testados
1Claude Fable 5
95.0
2Kimi K3
95.0
3GPT-5.6 Sol
93.1
4Claude Opus 4.8
87.5
5GPT-5.6 Terra
83.8
22 modelos testados
Arena
1Qwen3.8 Max (0902)
1671.2
2Claude Fable 5
1653.9
3MiMo-V2.6-Pro
1618.4
4GLM 5.3 Flash
1615.6
5GLM 5.2
1593.3
58 modelos testados
1Claude Fable 5
1507.6
2Claude Opus 4.6 (Fast)
1503.7
3Claude Opus 4.6
1497.3
4Claude Opus 4.7
1494.1
5
U
Muse Spark1489.3
148 modelos testados
Segurança
Matemática
FrontierMath-2025-02-28-Private
Matemática1
U
Muse Spark68.4
2Gemini 3 Pro
66.0
3GPT-5.4 Pro
50.0
4GPT-5.4
47.6
5Claude Opus 4.8
47.2
67 modelos testados
FrontierMath-Tier-4-2025-07-01-Private
Matemática1GPT-5.4 Pro
37.5
2GPT-5.2 Pro
31.3
3Claude Opus 4.8
31.3
4Gemini 3 Pro
31.3
5GPT-5.4
27.1
49 modelos testados
FrontierMath-Tier-4-v2-Private
Matemática1GPT-6.1 Sol
100.0
2GPT-6 Astra
97.6
3Claude Opus 5.5
95.0
4Claude Fable 5
90.2
5GPT-6 Sol
90.0
58 modelos testados
FrontierMath-Tiers-1-3-v2-Private
Matemática1GPT-6 Astra
93.7
2GPT-6.1 Sol
93.7
3Claude Opus 5.5
91.2
4Claude Fable 5.1
90.2
5GPT-6 Sol
89.8
77 modelos testados
GSM8K
Matemática1GPT-4 (older v0314)
92.0
2GPT-4o-mini (2024-07-18)
91.3
3GPT-4o-mini
91.3
4Qwen2.5 Coder 32B Instruct
91.1
5GPT-4 Turbo
90.0
43 modelos testados
HELM — Omni-MATH
Matemática1GPT-5 Mini
72.2
2o4 Mini
72.0
3o3
71.4
4gpt-oss-120b
68.8
5Kimi K2 0711
65.4
40 modelos testados
LiveBench — Mathematics
Matemática1DeepSeek V4 Pro
90.7
2GLM 5.2
89.8
3GPT-5.2-Codex
88.8
4Qwen3.7 Max
85.3
5GLM 5.1
84.9
40 modelos testados
MATH level 5
Matemática1GPT-5
98.1
2GPT-5 Mini
97.8
3o4 Mini
97.8
4o3
97.8
5Claude Sonnet 4.5
97.7
81 modelos testados
MATH Level 5
Matemática1Qwen2.5 32B Instruct
62.5
2
HA
Qwen2.5 72B Instruct Abliterated60.1
3Qwen2.5 72B Instruct
59.8
4DeepSeek R1 Distill Qwen 14B
57.0
5Qwen2.5 14B Instruct
55.3
78 modelos testados
OpenCompass — AIME2025
Matemática1DeepSeek V3.2 Speciale
96.0
2GLM 5
95.8
3Step 3.5 Flash
95.7
4GLM 4.7
95.4
5Kimi K2 Thinking
94.1
34 modelos testados
OTIS Mock AIME 2024-2025
Matemática1Claude Fable 5
100.0
2Claude Fable 5.1
100.0
3GPT-6 Astra
100.0
4Claude Opus 5.5
100.0
5GPT-6 Sol
100.0
168 modelos testados
Linguagem
HELM — IFEval
Linguagem1Grok 3 Mini Beta
95.1
2Grok-3 mini
95.1
3Grok 4
94.9
4GPT-5.1
93.5
5GPT-5 Nano
93.2
40 modelos testados
IFEval
Linguagem1Llama 3.3 70B Instruct
90.0
2Llama 3.1 70B Instruct
86.7
3Qwen2.5 72B Instruct
86.4
4
HA
Qwen2.5 72B Instruct Abliterated85.9
5Qwen2.5 32B Instruct
83.5
78 modelos testados
JCommonsenseQA
Linguagem1DeepSeek R1 Distill Qwen 32B
95.3
2DeepSeek R1 Distill Qwen 14B
93.7
3Qwen2 7B Instruct
89.1
4Qwen2 VL 7B Instruct
87.8
5Meta Llama 3 8B Instruct
87.7
12 modelos testados
JHumanEval
Linguagem0 modelos testados
JMMLU
Linguagem1DeepSeek R1 Distill Qwen 32B
72.2
2DeepSeek R1 Distill Qwen 14B
63.4
3Qwen2 7B Instruct
56.5
4Qwen2 VL 7B Instruct
56.3
5Meta Llama 3 8B Instruct
46.7
12 modelos testados
JNLI
Linguagem1DeepSeek R1 Distill Qwen 32B
85.6
2DeepSeek R1 Distill Qwen 14B
82.4
3Qwen2 7B Instruct
81.3
4Qwen2 VL 7B Instruct
74.4
5DeepSeek R1 Distill Llama 8B
69.4
12 modelos testados
JSQuAD
Linguagem1Qwen2 VL 7B Instruct
89.9
2DeepSeek R1 Distill Qwen 14B
89.8
3Qwen2 7B Instruct
89.6
4Meta Llama 3 8B Instruct
89.5
5DeepSeek R1 Distill Qwen 32B
89.2
12 modelos testados
LiveBench — If
Linguagem1Qwen3.7 Max
74.0
2GLM 5.1
68.5
3Gemma 4 31B
67.6
4GPT-5.1-Codex-Max
67.1
5GPT-5.2-Codex
66.5
40 modelos testados
LiveBench — Language
Linguagem1Qwen3.7 Max
79.7
2DeepSeek V4 Pro
78.1
3Kimi K2.7 Code
77.9
4GLM 5
77.5
5MiniMax M3
76.8
40 modelos testados
LLM-JP — Overall
Linguagem1DeepSeek R1 Distill Qwen 32B
59.3
2DeepSeek R1 Distill Qwen 14B
56.8
3Qwen2 VL 7B Instruct
53.0
4Qwen2 7B Instruct
51.7
5Meta Llama 3 8B Instruct
49.6
12 modelos testados
OpenCompass — IFEval
Linguagem1Kimi K2.5
93.9
2Step 3.5 Flash
93.2
3GLM 5
93.2
4Kimi K2 Thinking
92.4
5DeepSeek V3.2 Speciale
91.7
34 modelos testados
Multimodal
VideoMME
Multimodal1Gemini 1.5 Pro (Feb 2024)
66.7
2Qwen2.5 72B Instruct
64.7
3GPT-4o (2024-11-20)
62.5
4GPT-4o (2024-08-06)
62.5
5Gemini 1.5 Flash (May 2024)
60.4
8 modelos testados