AGIEval

Human-centric standardized exams (SAT, LSAT, Gaokao, GRE, GMAT, Civil Service, Lawyer Qualification).26 Oct 2026

One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
		Evidence	What the result shows
GPT-4o →OpenAI	69% 1 report	AGIEval · Accuracy · All · few-shot[1]AGIEval · All · few-shot	Leads this displayed comparison. Source notes Source cell: 69.0
GPT-3.5-Turbo →OpenAI	47.2% 1 report	AGIEval · Accuracy · All · few-shot[1]AGIEval · All · few-shot	#2 of 2 in this displayed comparison; 21.8 pp below the leader. Source notes Source cell: 47.2
GPT-4o →OpenAI	62.3% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	Leads this displayed comparison. Source notes Source cell: 62.3
InternLM2-20B* →	53% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#2 of 14 in this displayed comparison; 9.3 pp below the leader. Source notes Source cell: 53.0
Qwen-14B* →Qwen	52% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#3 of 14 in this displayed comparison; 10.3 pp below the leader. Source notes Source cell: 52.0
Phi-3-medium 14b* →Microsoft	50.2% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#4 of 14 in this displayed comparison; 12.1 pp below the leader. Source notes Source cell: 50.2
InternLM2-Chat-7B-SFT* →	49% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#5 of 14 in this displayed comparison; 13.3 pp below the leader. Source notes Source cell: 49.0
GPT-3.5-Turbo →OpenAI	46% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#6 of 14 in this displayed comparison; 16.3 pp below the leader. Source notes Source cell: 46.0
Qwen-7B* →Qwen	45.6% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#7 of 14 in this displayed comparison; 16.7 pp below the leader. Source notes Source cell: 45.6
Mixtral 8x7b* →Mistral	45.2% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#8 of 14 in this displayed comparison; 17.1 pp below the leader. Source notes Source cell: 45.2
Phi-3-small 7b* →Microsoft	45.1% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#9 of 14 in this displayed comparison; 17.2 pp below the leader. Source notes Source cell: 45.1
Gemma 7b* →Google	42.1% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#10 of 14 in this displayed comparison; 20.2 pp below the leader. Source notes Source cell: 42.1
Llama-3-In* →Meta	42% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#11 of 14 in this displayed comparison; 20.3 pp below the leader. Source notes Source cell: 42.0
Phi-3-mini 3.8b* →Microsoft	37.5% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#12 of 14 in this displayed comparison; 24.8 pp below the leader. Source notes Source cell: 37.5
Mistral 7b* →Mistral	35.1% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#13 of 14 in this displayed comparison; 27.2 pp below the leader. Source notes Source cell: 35.1
Phi-2 2.7b* →Microsoft	29.8% 1 report	AGIEval · Accuracy · All · zero-shot[1]AGIEval · All · zero-shot	#14 of 14 in this displayed comparison; 32.5 pp below the leader. Source notes Source cell: 29.8
GPT-4o →OpenAI	71.9% 1 report	AGIEval · Accuracy · Chinese · few-shot[1]AGIEval · Chinese · few-shot	Leads this displayed comparison. Source notes Source cell: 71.9
GPT-3.5-Turbo →OpenAI	49.5% 1 report	AGIEval · Accuracy · Chinese · few-shot[1]AGIEval · Chinese · few-shot	#2 of 2 in this displayed comparison; 22.4 pp below the leader. Source notes Source cell: 49.5
GPT-4o →OpenAI	63.3% 1 report	AGIEval · Accuracy · Chinese · zero-shot[1]AGIEval · Chinese · zero-shot	Leads this displayed comparison. Source notes Source cell: 63.3
GPT-3.5-Turbo →OpenAI	45% 1 report	AGIEval · Accuracy · Chinese · zero-shot[1]AGIEval · Chinese · zero-shot	#2 of 2 in this displayed comparison; 18.3 pp below the leader. Source notes Source cell: 45.0
GPT-4o →OpenAI	71.4% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	Leads this displayed comparison. Source notes Source cell: 71.4
Llama 3 400B+ →Meta	69.9% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#2 of 8 in this displayed comparison; 1.5 pp below the leader. Source notes Source cell: 69.9
Llama 3 70B →Meta	63% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#3 of 8 in this displayed comparison; 8.4 pp below the leader. Source notes Source cell: 63
Mixtral 8x22B →Mistral	61.2% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#4 of 8 in this displayed comparison; 10.2 pp below the leader. Source notes Source cell: 61.2
GPT-3.5-Turbo →OpenAI	52.7% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#5 of 8 in this displayed comparison; 18.7 pp below the leader. Source notes Source cell: 52.7
Llama 3 8B →Meta	45.9% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#6 of 8 in this displayed comparison; 25.5 pp below the leader. Source notes Source cell: 45.9
Gemma 7B →Google	44.9% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#7 of 8 in this displayed comparison; 26.5 pp below the leader. Source notes Source cell: 44.9
Mistral 7B →Mistral	44% 1 report	AGIEval · Accuracy · English · few-shot[1]AGIEval · English · few-shot	#8 of 8 in this displayed comparison; 27.4 pp below the leader. Source notes Source cell: 44
GPT-4o →OpenAI	65.2% 1 report	AGIEval · Accuracy · English · zero-shot[1]AGIEval · English · zero-shot	Leads this displayed comparison. Source notes Source cell: 65.2
GPT-3.5-Turbo →OpenAI	54.1% 1 report	AGIEval · Accuracy · English · zero-shot[1]AGIEval · English · zero-shot	#2 of 2 in this displayed comparison; 11.1 pp below the leader. Source notes Source cell: 54.1

Sources

AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models — published result table ↗
Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang et al. · 2023
All 9 authors
Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen, Nan Duan
Dates & evaluation details
AGIEval · Accuracy · English · few-shot · 2026-10-26
Published model comparison, 2023. AGIEval · English · few-shot. Accuracy · English · few-shot; AGIEval · English · few-shot. The table retains the source’s reported model and evaluation configuration. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
AGIEval · Accuracy · Chinese · few-shot · 2026-10-26
Published model comparison, 2023. AGIEval · Chinese · few-shot. Accuracy · Chinese · few-shot; AGIEval · Chinese · few-shot. The table retains the source’s reported model and evaluation configuration. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
AGIEval · Accuracy · All · few-shot · 2026-10-26
Published model comparison, 2023. AGIEval · All · few-shot. Accuracy · All · few-shot; AGIEval · All · few-shot. The table retains the source’s reported model and evaluation configuration. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
AGIEval · Accuracy · English · zero-shot · 2026-10-26
Published model comparison, 2023. AGIEval · English · zero-shot. Accuracy · English · zero-shot; AGIEval · English · zero-shot. The table retains the source’s reported model and evaluation configuration. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
AGIEval · Accuracy · Chinese · zero-shot · 2026-10-26
Published model comparison, 2023. AGIEval · Chinese · zero-shot. Accuracy · Chinese · zero-shot; AGIEval · Chinese · zero-shot. The table retains the source’s reported model and evaluation configuration. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
AGIEval · Accuracy · All · zero-shot · 2026-10-26
Published model comparison, 2023. AGIEval · All · zero-shot. Accuracy · All · zero-shot; AGIEval · All · zero-shot. The table retains the source’s reported model and evaluation configuration. Published result table; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

What it measures

Human-centric standardized exams (SAT, LSAT, Gaokao, GRE, GMAT, Civil Service, Lawyer Qualification).

Reading the measurement

Study what information the model receives and which inference the answer requires. A question may mix recall, visual perception, deduction and search. Use task breakdowns to find the mechanism behind a headline gain.

Published evaluation

Accuracy · English · few-shot; AGIEval · English · few-shot. The table retains the source’s reported model and evaluation configuration.

Where this fits in an evaluation suite

Ask what the task actually requires: recall, deduction, search, or a mixture. Separate closed-book reasoning from retrieval and tool-assisted problem solving.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Accuracy · English · few-shot; AGIEval · English · few-shot. The table retains the source’s reported model and evaluation configuration.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models — published result table ↗
Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang et al. · 2023
All 9 authors
Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen, Nan Duan

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All reasoning benchmarks →

Sources

What it measures

Reading the measurement

Published evaluation

Where this fits in an evaluation suite

A reproducible comparison

Continue exploring

Artificial Analysis Intelligence Index v5.0

Humanity’s Last Exam

MMMU-Pro

ARC-AGI-2