IBM logo

IBM Granite 4.2 3B

IBM

IBM Granite 4.2 3B is the compact 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. Unlike the 8B and 30B, it did not receive the extra agentic RL block. Open weights on Hugging Face under Apache 2.0. Text-only (not multimodal). It is served by deepinfra with a 128K-token context window at $0.03 / $0.12 per 1M input/output tokens.

Key Specifications

Parameters
3.0B
Context
131.1K
Release Date
August 25, 2026
Average Score
55.7%

Timeline

Key dates in the model's history
Announcement
August 25, 2026
Last Update
September 12, 2026
Today
September 20, 2026

Technical Specifications

Parameters
3.0B
Training Tokens
-
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.03
Output (per 1M tokens)
$0.12
Max Input Tokens
131.1K
Max Output Tokens
131.1K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Reasoning

Logical reasoning and analysis
GPQA
ScoreSelf-reported
54.8%

Other Tests

Specialized benchmarks
AIME 2025
ScoreSelf-reported
78.3%
Arena-Hard v2
ScoreSelf-reported
35.0%
BFCL-V4
ScoreSelf-reported
52.4%
HMMT25
ScoreSelf-reported
66.7%
IFBench
ScoreSelf-reported
74.3%
LiveCodeBench v6
ScoreSelf-reported
69.7%
MMLU-Pro
ScoreSelf-reported
67.8%
ProfBench
ScoreSelf-reported
32.1%
RULER 128k
ScoreSelf-reported
55.3%
RULER 64k
ScoreSelf-reported
67.5%
SciCode
ScoreSelf-reported
24.1%
TAU3-Bench
ScoreSelf-reported
45.8%

License & Metadata

License
apache_2_0
Announcement Date
August 25, 2026
Last Updated
September 12, 2026

Compare IBM Granite 4.2 3B

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.