Model Details
IBM Granite 4.0 Tiny Preview
Quick answer: IBM Granite 4.0 Tiny Preview is IBM's June 2025 ultra-compact model scoring 82.4% HumanEval, 70.1% GSM8K, and 26.7% Arena Hard. Apache 2.0 — "Tiny" tier in the Granite 4.0 preview family.
At a Glance
Where Granite 4.0 Tiny Preview leads
- 82.4% HumanEval — excellent coding for Tiny scale
- Apache 2.0 — fully open
- IBM enterprise integration
Where it lags
- 26.7% Arena Hard — limited instruction quality
- 70.1% GSM8K — moderate math
- Preview: not production-ready
Best for: IBM enterprise compact coding tasks; Apache 2.0 tiny-scale HumanEval; Granite ecosystem.
What IBM Granite 4.0 Tiny Preview Is
IBM Granite 4.0 Tiny Preview is the smallest model in IBM's June 2025 Granite 4.0 preview release — a compact model for minimal hardware deployments with IBM enterprise focus.
Specifications
| Field | Value |
|---|---|
| Organization | IBM |
| License | Apache 2.0 |
| HuggingFace | ibm-granite/granite-4.0-tiny-preview |
| Release date | June 2025 |
| Modality | Text only |
Pricing
Open weights under Apache 2.0 — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| HumanEval | 82.4% | Benchgen evaluation | 2025-06 |
| GSM8K | 70.1% | Benchgen evaluation | 2025-06 |
| Arena Hard | 26.7% | Benchgen evaluation | 2025-06 |
Granite 4.0 Tiny vs Alternatives
| Model | HumanEval | GSM8K | License |
|---|---|---|---|
| Granite 4.0 Tiny Preview | 82.4% | 70.1% | Apache 2.0 |
| Granite 3.3 8B Instruct | — | — | Apache 2.0 |
| Phi-3.5 Mini Instruct | — | 86.2% | Apache 2.0 |
For IBM enterprise compact coding: Granite 4.0 Tiny. For math-focused tiny: Phi-3.5 Mini (86.2% GSM8K).
Related Models & Benchmarks
- Compare: Granite 3.3 8B Instruct · Phi-3.5 Mini Instruct
- Benchmarks: HumanEval · GSM8K · Arena Hard
Frequently Asked Questions
What is IBM Granite 4.0 Tiny Preview?
IBM's June 2025 tiny preview model scoring 82.4% HumanEval, 70.1% GSM8K. Apache 2.0 — compact enterprise coding.Specs from IBM's Granite 4.0 Tiny Preview release (June 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.