Granite 3.3 8B (Non-reasoning)

Granite 3.3 8B (Non-reasoning)

IBM · Released Apr 16, 2025 Efficient
Intelligence #303 / 599
30.3 Our Score
Speed #6 / 281
359.8 tokens / sec
Input #143 / 599
$0.030 per 1M tokens
Output #205 / 599
$0.250 per 1M tokens
Context
— Not reported

Analysis Summary

Granite 3.3 8B Non-reasoning is IBM's compact open-weight model, priced at just $0.03 input and $0.25 output per million tokens. Intelligence index sits at 1.8 with MMLU-Pro of 0.468, placing it in the lower tier, but the agentic index of 10.5 and tau2 score of 0.105 suggest some structured task-following capability beyond raw reasoning.

For businesses, its primary appeal is cost efficiency for high-volume, low-complexity tasks such as classification, simple extraction, and lightweight tool-use pipelines. Coding capability is limited at LiveCodeBench 0.127, and long-context reliability (LCR 0.043) is weak, so it is not suited to document-heavy or engineering workflows.

At this price point it is one of the most affordable options in the database. Teams running large-scale automation on a tight budget will find it useful for narrow, well-defined tasks, but should pair it with a stronger model for anything requiring reasoning depth.

Assessed June 30, 2026

Editorial notes

Granite 3.3 8B Non-reasoning from IBM is an ultra-cheap small model with strong value credentials, basic agentic capability, and tool-use potential, though reasoning and coding benchmarks are limited.

Rankings consider pricing, capabilities, benchmarks, and real-world applicability and are refreshed as new models launch. Feedback?

Performance Profile

Intelligence1.1Technical1.3Value7.3Content3
Intelligence 1.1/10
Technical 1.3/10
Content 3/10
Value 7.3/10

How Granite 3.3 8B (Non-reasoning) compares

Granite 3.3 8B (Non-reasoning) ranks #374 of 387 AI models we track for overall intelligence, #259 of 294 for agentic tasks. At $0.03 per million input tokens it is cheaper than 76% of comparable models.

Performance Indices

Source: Artificial Analysis

1.8 Intelligence Index
10.5 Agentic Index
6.7 Math Index

Benchmark Scores

Intelligence

GPQA Diamond 33.8% Graduate-level scientific reasoning
HLE 4.2% Humanity's Last Exam
MMLU Pro 46.8% Multi-task language understanding
MATH 500 66.5% Mathematical problem-solving
AIME 4.7% Competition mathematics
AIME 2025 6.7% Competition mathematics (2025)
SciCode 10.1% Scientific computing

Technical

LiveCodeBench 12.7% Live coding evaluation
τ²-Bench 10.5% Conversational agent benchmark

Content

IFBench 22.4% Instruction following
LCR 4.3% Long-context reasoning

Benchmark data from Artificial Analysis and Hugging Face

How does Granite 3.3 8B (Non-reasoning) stack up?

Compare side-by-side with other efficient models.

Compare Models

Model Information

ProviderIBM
Release Date April 16, 2025
Status Active

Pricing

Token Type Cost per 1M tokens Cost per 1K tokens
Input $0.03 $0.000030
Output $0.25 $0.000250

Leaderboard Categories

Frequently asked questions about Granite 3.3 8B (Non-reasoning)

How much does Granite 3.3 8B (Non-reasoning) cost?

Granite 3.3 8B (Non-reasoning) costs $0.03 per million input tokens and $0.25 per million output tokens.

Who created Granite 3.3 8B (Non-reasoning)?

Granite 3.3 8B (Non-reasoning) is developed by IBM and was released on April 16, 2025.