Inference
Epoch AI

MATH level 5

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.88
Top resultBest result on this test.98.1%GPT-5
Top-3 spreadPoints from first to third.0.3 pts
YearYear of release.N/A

Result and price

020406080100$0.01$0.03$0.1$0.3$1$3$10$30$100RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

88 results
#ModelResultThe score from the source.
01GPT-5high effort98.1% ±0.3
02GPT-5 minihigh effort97.8% ±0.3
03o4-mini (high)high effort97.8% ±0.3
04o3high effort97.8% ±0.3
05Claude Sonnet 4.597.7% ±0.4
06Qwen3 Max97.1% ±0.4
07DeepSeek-R1 (May 2025)96.6% ±0.4
08o3-minihigh effort96.5% ±0.4
09Gemini 2.5 Pro (May 2025)95.9% ±0.4
10Gemini 2.5 Pro (Mar 2025)95.6% ±0.4
11GPT-5 nanohigh effort94.9% ±0.5
12o1high effort94.7% ±0.6
13DeepSeek-R193.1% ±0.7
14Claude Haiku 4.591.6% ±0.5
15R1 Distill Llama 70B89.9% ±0.7
16O1 Minihigh effort89.2% ±0.7
17Grok 3 [Beta]88.7% ±0.9
18Claude 3.7 Sonnet88.1% ±0.8
19Grok 3 Minihigh effort88.1% ±0.9
20GPT-4.1 mini87.3% ±0.7
21DeepSeek R1 Distill Qwen 14B87.1% ±0.7
22Claude Opus 485.0% ±1.0
23Claude Sonnet 484.4% ±1.0
24Gemini 2.0 Pro83.5% ±1.0
25GPT-4.183.0% ±0.8
26Gemini 2.0 Flash82.2% ±0.9
27o1-preview81.6% ±1.1
28Mistral Medium 381.6% ±0.9
29GPT-4.578.6% ±1.1
30DeepSeek-V3 (Mar 2025)75.5% ±0.9
31Gemma 3 27B74.0% ±1.1
32Llama 4 Maverick73.0% ±1.1
33GPT-4.1 nano70.0% ±1.0
34Qwen3 235B A22B68.9% ±0.9
35Qwen2.5 Max67.2% ±1.1
36Qwen Plus 072865.3% ±1.1
37Phi-464.9% ±1.1
38DeepSeek V364.9% ±1.0
39Grok 263.5% ±1.3
40Qwen2.5-72B63.2% ±1.1
41Llama 4 Scout62.3% ±1.2
42Claude 3.5 Sonnet (October 2024)56.9% ±1.1
43Qwen Turbo56.2% ±1.1
44Qwen2.5 32B56.1% ±1.1
45Gemini 1.5 Pro55.6% ±1.1
46GPT-4o mini52.6% ±1.1
47Claude 3.5 Sonnet51.7% ±1.2
48GPT-4o51.0% ±1.1
49Llama 3.1 405B49.8% ±1.2
50Mistral Large 247.6% ±1.1
51Mistral Small 3.146.8% ±1.2
52GPT-4 Turbo (Apr 2024)46.7% ±1.1
53Claude 3.5 Haiku46.4% ±1.2
54Mistral Small 344.8% ±1.1
55Gemini 1.5 Flash43.5% ±1.1
56Tulu 3 (Tülu 3) 70B42.7% ±1.0
57Llama 3.3 70B41.6% ±1.2
58Llama 3.2 90B39.4% ±1.0
59Qwen2 72B39.1% ±1.0
60GPT-4 Turbo (Nov 2023)37.7% ±1.1
61Claude 3 Opus37.5% ±1.1
62Llama 3.1-70B36.7% ±1.0
63Gemma 2 27B27.9% ±1.0
64WizardLM-2 8x22B25.7% ±0.9
65Yi 1.5 34B25.5% ±0.9
66Mistral Large24.5% ±0.9
67Mixtral 8x22B24.2% ±0.8
68GPT-4 (Jun 2023)23.0% ±0.8
69Llama 3.1-8B22.9% ±0.9
70Hermes 2 Theta Llama-3 70B22.7% ±0.8
71Llama 3 70B22.6% ±0.8
72Gemma 2 9B21.0% ±0.9
73Claude 3 Sonnet18.2% ±0.8
74phi-3-medium 14B17.6% ±0.7
75Ministral 8B14.9% ±0.7
76Claude 3 Haiku14.9% ±0.7
77Ministral 3B14.4% ±0.7
78GPT-3.5 Turbo13.8% ±0.6
79Claude 211.7% ±0.7
80DBRX11.7% ±0.7
81Gemini 1.0 Pro11.2% ±0.6
82Mistral Nemo10.8% ±0.6
83Mixtral 8x7B9.6% ±0.5
84DeepSeek LLM 67B6.4% ±0.4
85Llama 3-8B6.1% ±0.4
86Yi 34B5.1% ±0.3
87Mistral 7B3.6% ±0.3
88Llama 2-70B3.3% ±0.3