跳到正文
智迹闻AuraTracer

发布于

Granite 4.2 3B

公开成绩与计算依据

综合参考分根据已公开的评测资料整理。查看发布日期依据

综合参考分 31.3 · 20项基准

推理求解 32.5

AIME · 83.5

huggingface.co · 官方发布比较 · 2026

granite-4.2-3B · Official published comparison; AIME 2026; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

MATH-500 · 97.0

huggingface.co · 官方发布比较 · 500-item subset

granite-4.2-3B · Official published comparison; MATH-500; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

GPQA Diamond · 0.558585858585859 比例

Artificial Analysis · 独立评测 · Diamond

Granite 4.2 3B · Current AA standalone protocol. Status verified in aa-method-current.md section gpqa-diamond

Terminal-Bench-Science · 0 比例

Artificial Analysis · 独立评测 · 0.1.0

Granite 4.2 3B · Artificial Analysis current published protocol; terminal-bench-science

HLE · 0.029 比例

Artificial Analysis · 独立评测 · Diamond

Granite 4.2 3B · Artificial Analysis current published protocol; hle-diamond

CritPt · 0 比例

Artificial Analysis · 独立评测 · 2 Hard

Granite 4.2 3B · Artificial Analysis current published protocol; critpt-v2-hard

工具与任务执行 9.8

tau3-Banking · 0.0556701030927835 比例

Artificial Analysis · 独立评测 · 1.0.1

Granite 4.2 3B · Current AA standalone protocol. Status verified in aa-method-current.md section tau3-banking

GDPval-AA · 209.24 分

Artificial Analysis · 独立评测 · 2.1

Granite 4.2 3B · Artificial Analysis current published protocol; gdpval-aa-normalized

Terminal-Bench · 0 比例

Artificial Analysis · 独立评测 · 4.0

Granite 4.2 3B · Artificial Analysis current published protocol; terminalbench-4-0

AutomationBench · 0.003767979122295127 比例

Artificial Analysis · 独立评测 · current published task suite

Granite 4.2 3B · Artificial Analysis current published protocol; automationbench-aa

AA-Briefcase · 98.83 分

Artificial Analysis · 独立评测 · 1.1

Granite 4.2 3B · Artificial Analysis current published protocol; aa-briefcase

BFCL · 52.2

huggingface.co · 官方发布比较 · 4

granite-4.2-3B · Official published comparison; BFCL v4; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

编程 34.9

FrontierCode · 0 比例

Artificial Analysis · 独立评测 · 2.0

Granite 4.2 3B · Artificial Analysis current published protocol; frontiercode-2-0

LiveCodeBench · 58.9

huggingface.co · 官方发布比较 · release_v6

granite-4.2-3B · Official published comparison; LiveCodeBench v6; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

语言与信息处理 52.9

IFBench · 73.0

huggingface.co · 官方发布比较 · current published task suite

granite-4.2-3B · Official published comparison; IFBench; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

Multi-IF · 75.9

huggingface.co · 官方发布比较 · current maintainer dataset

granite-4.2-3B · Official published comparison; Multi-IF; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

GDP.pdf · 0.002 比例

Artificial Analysis · 独立评测 · current published task suite

Granite 4.2 3B · Artificial Analysis current published protocol; gdp-pdf

IFEval · 93.7

huggingface.co · 官方发布比较 · current maintainer dataset

granite-4.2-3B · Official published comparison; IFEval; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

MMLU-Pro · 65.8

huggingface.co · 官方发布比较 · current maintainer dataset

granite-4.2-3B · Official published comparison; MMLU-Pro; exact column granite-4.2-3B. Footnotes and all competing runs retained in source.

AA-Omniscience · 0.092 比例

Artificial Analysis · 独立评测 · 1.1

Granite 4.2 3B · Artificial Analysis current published protocol; omniscience-accuracy

AA-Omniscience · 0.6967694566813509 比例

Artificial Analysis · 独立评测 · 1.1

Granite 4.2 3B · Artificial Analysis current published protocol; omniscience-non-hallucination

查看原始成绩与计算数据

资料来源

模型资料来源 · 目录许可