Granite 4.2 3B

Name: Granite 4.2 3B
Author: IBM

IBM Released August 2026

Full benchmarks

#87 smartest of 89 Updated Oct 31, 2026, 06:59 PM UTC

At a glance

How Granite 4.2 3B compares with the other 88 ranked language models.

Better for simple tasks. Ranked #87 of 89 on the overall intelligence score, ahead of 2% of models.
Answers right away. You wait about 0.00 sec before the answer starts.

Model details

Made by: IBM
Model family: Granite 4.2 3B
Released: August 2026 · 2 months ago
Input price: $0.00 / 1M tokens
Output price: $0.00 / 1M tokens
Value for money: –
Tests with scores: 6 of 8

Good for

No standout strengths in the tests.

Not ideal for

Long answers in real timeSlower than 90% of models
Complex reasoningScores in the bottom quarter overall
Autonomous agentsWeak on multi-step computer tasks

Intelligence

overall score

#87 of 89typical 30

Coding

–

coding score

typical –

Speed

tokens per second

#80 of 89typical 91

Wait

0.00 sec

before it replies

typical 16s

Price

–

per 1M tokens

typical $2.00

Test results

Scores on individual tests that make up the intelligence score. The faint line marks the best score of any model.

Humanity's Last Exam
Very hard expert questions across many subjects
7%#85 of 86typical 42%
GPQA Diamond
PhD-level biology, physics and chemistry questions
56%#35 of 36typical 90%
SciCode
Writing code to solve science research problems
25%#85 of 86typical 53%
Long documents
Finding and reasoning over facts in very long texts
24%#86 of 87typical 80%
Terminal-Bench
Doing real tasks on a computer by itself, like an agent
0%#84 of 89typical 31%
Customer support
Helping bank customers as a support agent using tools
6%#33 of 34typical 39%

Where it ranks on intelligence

This model next to the models closest to it. Click another bar to open that model.

What would it cost?

Through the API, at the listed price. Chat apps usually charge a monthly plan instead.

1,000 chat replies
a question and a ~350-word answer, 1,000 times
Free
Summarise a 100-page report
about 50,000 words in, a one-page summary out
Free
Write a 2,000-word article
a short brief in, a long article out
Free

Input, per 1M tokens

$0.00

Output, per 1M tokens

$0.00

How long do answers take?

Median times measured through the API.

Before the answer starts0.00 sec

Writing speed–

A full 500-word answer–

1 token is about ¾ of a word. Speeds change with server load, so treat these as typical, not guaranteed.

Similar models

Other companies' models with the closest intelligence score.

Questions people ask

Is Granite 4.2 3B a good AI model?

On the overall intelligence score it ranks #87 of 89 (score 5; the typical model scores 30).

How much does Granite 4.2 3B cost?

Through the API it costs $0.00 per million input tokens and $0.00 per million output tokens. In everyday terms, 1,000 chat replies cost about Free. Chat apps that use it usually charge a monthly plan instead.

How fast is Granite 4.2 3B?

It writes about – words per second, and you wait about 0.00 sec before the answer starts. A full 500-word answer takes roughly –.

What are good alternatives to Granite 4.2 3B?

Models with a similar score include Mistral Small 4 by Mistral, Nemotron 3.5 Lightning by NVIDIA, Command A+ by Cohere.

Benchmark data from Artificial Analysis, an independent benchmarking company. Rankings refresh every few hours.