Granite 4.2 8B
IBM开源权重Apache 2.0 · 商用许可
描述
IBM Granite 4.2 8B is a mid-size dense decoder-only reasoning LLM in the Granite 4.2 family. It supports native think, non-think, and low-effort modes, tool calling, and a 128K native context window with extension to 512K. Like the 30B, it received additional agentic reinforcement learning for multi-step tool use. Open weights on Hugging Face under Apache 2.0. Text-only (not multimodal).
发布日期
2026-08-25
参数规模
8.0B
上下文长度
131K
支持模态
text
能力雷达图
11
general
24
coding
63
reasoning
41
science
40
agents
0
multimodal
排行榜排名
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))Agents
TAU3-Bench
58.1%自报
General
Arena-Hard v2
65.2%自报
Instruction Following
IFBench
79.3%自报
Language
MMLU-Pro
74.0%自报
Long Context
RULER 64k
81.0%自报
RULER 128k
71.4%自报
Math
AIME 2025
86.7%自报
HMMT25
78.3%自报
Reasoning
LiveCodeBench v6
73.2%自报
GPQANYU + Cohere + Anthropic (2023)
64.1%自报
SWE-Bench Verified
47.7%自报
ProfBench
41.2%自报
Bird-SQL (dev)
41.1%自报
SciCode
36.1%自报
SWE-bench Multilingual
30.8%自报
Terminal-Bench 2.1
20.6%自报
SWE-Bench ProPrinceton NLP (2024)
19.1%自报
Tool Calling
BFCL-V4
52.4%自报
AA 评测指数
(Artificial Analysis)Gpqa(NYU + Cohere + Anthropic (2023))63.1
Lcr(Artificial Analysis)45.0
Scicode(UIUC + Argonne National Lab (2024))31.5
Coding Index(Artificial Analysis)22.4
Terminalbench V2 118.4
Intelligence Index(Artificial Analysis)11.1
Hle(Center for AI Safety + Scale AI (2025))9.7
Tau Banking7.6
Terminalbench V4 00.0
LLM Stats 分类评分
(LLM Stats (zeroeval))Instruction Following80
Long Context80
Language70
Legal70
Math70
Finance70
Healthcare70
Creativity70
Writing70
General60
Physics50
Reasoning50
Frontend Development50
Biology50
Chemistry50
Knowledge40
Agents40
Tool Calling40
Code30
定价
输入价格$0.06 / 1M tokens
输出价格$0.25 / 1M tokens
混合价格(3:1)$0.107 / 1M tokens
缓存读取价格$0.015 / 1M tokens
速度
Tokens/秒40.2
首Token延迟0.42s
首回答延迟50.16s
供应商价格排行
供应商价格排行
6 个供应商
最便宜: DeepInfra最贵: Kilo Gateway
供应商输入输出
1DeepInfra最便宜
$0
$0
2IBM主要
$0.06
$0.25
3NanoGPT
$0.06
$0.15
4OpenRouter
$0.06
$0.25
5CoreWeave
$0.06
$0.15
6Kilo Gateway
$0.06
$0.15
比较该模型在不同 API 供应商之间的定价。