BIG-Bench Hard

Models scored

88

Best score

0.89

Metric

Average

BIG-Bench Hard is an AI evaluation tracked by Epoch AI, with 88 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 February 2023 to December 2024. The best score recorded is 0.89, measured as Average. Models come from Alibaba, Google DeepMind, Mistral AI, Meta AI, DeepSeek, 01.AI, Baichuan, NVIDIA and others.

Full record

Score metric: Average
Models scored: 88
Best score recorded: 0.89
Earliest model scored: Feb 24, 2023
Most recent model scored: Dec 26, 2024
Third-party reported: Yes

Leaderboard

01	gemini-1.5-pro-001	Google DeepMind	0.89
02	DeepSeek-V3	DeepSeek	0.88
03	gemini-1.5-pro-001-feb24	Google DeepMind	0.84
04	Llama-3.1-405B	Meta AI	0.83
05	Phi-3-medium-128k-instruct	Microsoft	0.81
06	Qwen2.5-72B	Alibaba	0.8
07	Phi-3-small-8k-instruct	Microsoft	0.79
08	DeepSeek-V2	DeepSeek	0.79
09	gpt-4-0613	OpenAI	0.75
10	Yi-34B-Chat	01.AI	0.72
11	Phi-3-mini-4k-instruct	Microsoft	0.72
12	StableBeluga2	Stability AI	0.69
13	Llama-2-70b-hf	Meta AI	0.65
14	gpt-3.5-turbo-0613	OpenAI	0.62
15	phi-2	Microsoft	0.59
16	Nemotron-4 15B	NVIDIA	0.59
17	Llama-2-70b-chat	Meta AI	0.58
18	LLaMA-65B	Meta AI	0.58
19	Llama-2-13b-chat	Meta AI	0.58
20	Mistral-7B-v0.1	Mistral AI	0.56
21	gemma-7b	Google DeepMind	0.55
22	Qwen-14B-Chat	Alibaba	0.55
23	Yi-34B	01.AI	0.54
24	Qwen-14B	Alibaba	0.53
25	internlm-20b		0.53
26	LLaMA-33B	Meta AI	0.5
27	Baichuan-2-13B-Base	Baichuan	0.49
28	Baichuan2-13B-Chat		0.47
29	Yi-6B-Chat	01.AI	0.47
30	Llama-2-13b	Meta AI	0.47
31	Qwen-7B	Alibaba	0.45
32	Llama-2-34b	Meta AI	0.44
33	vicuna-13b-v1.1		0.43
34	Baichuan-13B-Base	Baichuan	0.43
35	Yi-6B	01.AI	0.43
36	internlm-chat-20b		0.42
37	Baichuan-2-7B-Base	Baichuan	0.42

More benchmarks

Epoch Capabilities Index MMLU Terminal-Bench GPQA Diamond OTIS Mock AIME 2024-2025 GSM8K ARC-AGI ARC-AGI-2

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