Balrog — Epoch AI

Balrog — Epoch AI is run by Epoch AI. It has ranked 36 models, of which the catalogue holds 28, scoring from 0.037 to 0.581.

measured by Epoch AI · the board itself

Place	Model	Metric	Score
1stof 36	gemini-3-pro-preview	Average progress	0.581
2ndof 36	Gemini 3.1 Pro Preview	Average progress	0.57
4thof 36	Gemini 3 Flash Preview	Average progress	0.481
6thof 36	Claude Opus 4.5	Average progress	0.435
10thof 36	DeepSeek-R1	Average progress	0.349
11thof 36	Gemini 2.5 Flash	Average progress	0.335
12thof 36	GPT-5	Average progress (minimal)	0.328
13thof 36	Claude 3.5 Sonnet	Average progress	0.326
14thof 36	GPT-4o (2024-05-13)	Average progress	0.323
15thof 36	Claude Haiku 4.5	Average progress	0.312
17thof 36	Grok 3	Average progress	0.295
18thof 36	Reka Flash 3	Average progress	0.292
19thof 36	Meta-Llama-3.1-70B-Instruct	Average progress	0.279
20thof 36	Llama-3.2-90B-Vision-Instruct	Average progress	0.273
21stof 36	Llama 3.3 70B Instruct	Average progress	0.23
23rdof 36	DeepSeek R1 Distill QWEN 32B	Average progress	0.195
24thof 36	Claude Haiku 3.5	Average progress	0.193
25thof 36	Mistral Nemo	Average progress	0.176
26thof 36	GPT-4o-mini (2024-07-18)	Average progress	0.174
27thof 36	Llama 3.2 11B Vision Instruct	Average progress	0.168
28thof 36	Qwen2.5 72B Instruct	Average progress	0.162
29thof 36	Llama 3.1 8B Instruct	Average progress	0.151
31stof 36	Qwen 2 VL 72B Instruct	Average progress	0.128
32ndof 36	Phi-4	Average progress	0.116
33rdof 36	Llama 3.2 3B Instruct	Average progress	0.101
34thof 36	Qwen2.5 7B Instruct	Average progress	0.078
35thof 36	Llama 3.2 1B Instruct	Average progress	0.066
36thof 36	Qwen2-VL-7B-Instruct	Average progress	0.037

Place

Model

Metric

Score

1stof 36

Average progress

0.581

2ndof 36

Average progress

0.57