Benchmarks
Balrog
Models scored
36
Best score
0.58
Metric
Average progress
Balrog is an AI evaluation tracked by Epoch AI, with 36 scored model versions on record. The scores here are reported by third parties rather than produced by Epoch AI running the evaluation itself.
Scored models span May 2024 to February 2026. The best score recorded is 0.58, measured as Average progress. Models come from Google DeepMind, xAI, Anthropic, DeepSeek, OpenAI, Reka AI, Meta AI, Mistral AI and others.
Full record
- Score metric
- Average progress
- Models scored
- 36
- Best score recorded
- 0.58
- Earliest model scored
- May 13, 2024
- Most recent model scored
- Feb 19, 2026
- Third-party reported
- Yes
Leaderboard
| 01 | gemini-3-pro-preview | Google DeepMind | 0.58 |
| 02 | gemini-3.1-pro-preview | Google DeepMind | 0.57 |
| 03 | gemini-3-flash-preview | Google DeepMind | 0.48 |
| 04 | grok-4-0709 | xAI | 0.44 |
| 05 | claude-opus-4-5-20251101_unknown | Anthropic | 0.43 |
| 06 | claude-opus-4-5-20251101 | Anthropic | 0.43 |
| 07 | gemini-2.5-pro-exp-03-25 | Google DeepMind | 0.43 |
| 08 | claude-opus-4-5-20251101_64K | Anthropic | 0.43 |
| 09 | DeepSeek-R1 | DeepSeek | 0.35 |
| 10 | gemini-2.5-flash | Google DeepMind | 0.34 |
| 11 | gpt-5-2025-08-07_minimal | OpenAI | 0.33 |
| 12 | claude-3-5-sonnet-20241022 | Anthropic | 0.33 |
| 13 | gpt-4o-2024-05-13 | OpenAI | 0.32 |
| 14 | claude-haiku-4-5-20251001_1K | Anthropic | 0.31 |
| 15 | claude-haiku-4-5-20251001 | Anthropic | 0.31 |
| 16 | grok-3-beta | xAI | 0.29 |
| 17 | reka-flash-3 | Reka AI | 0.29 |
| 18 | Llama-3.1-70B-Instruct | Meta AI | 0.28 |
| 19 | Llama-3.2-90B-Vision-Instruct | Meta AI | 0.27 |
| 20 | Llama-3.3-70B-Instruct | Meta AI | 0.23 |
| 21 | gemini-1.5-pro-002 | Google DeepMind | 0.21 |
| 22 | DeepSeek-R1-Distill-Qwen-32B | DeepSeek | 0.2 |
| 23 | claude-3-5-haiku-20241022 | Anthropic | 0.19 |
| 24 | Mistral-Nemo-Instruct-2407 | Mistral AI | 0.18 |
| 25 | gpt-4o-mini-2024-07-18 | OpenAI | 0.17 |
| 26 | Llama-3.2-11B-Vision-Instruct | Meta AI | 0.17 |
| 27 | qwen2.5-72b-instruct | Alibaba | 0.16 |
| 28 | Llama-3.1-8B-Instruct | Meta AI | 0.15 |
| 29 | gemini-1.5-flash-002 | Google DeepMind | 0.15 |
| 30 | Qwen2-VL-72B-Instruct | 0.13 | |
| 31 | phi-4 | Microsoft Research | 0.12 |
| 32 | Llama-3.2-3B-Instruct | Meta AI | 0.1 |
| 33 | qwen2.5-7b-instruct | Alibaba | 0.08 |
| 34 | Llama-3.2-1B-Instruct | Meta AI | 0.07 |
| 35 | Qwen2-VL-7B-Instruct | 0.04 |
More benchmarks
SourceEpoch AI, 'AI Benchmarking Hub'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/benchmarks. Licensed under CC BY 4.0.
← All benchmarks