benchmarks
IBM Granite 4.0 Micro is a 3-billion-parameter, text-only instruction model available under the Apache 2.0 license, with a 131,072-token context window. It is designed for enterprise language tasks, including retrieval-augmented generation, summarization, information extraction and tool calling.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 28.3%30 Aug 2026 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 2.8%30 Aug 2026 | View |
Weighted mean of 2 verified benchmark results: GPQA Diamond 28.3%, OTIS Mock AIME 2024-25 2.8%.
Capabilities
- Retrieval-augmented question answering over supplied documents.
- Text summarization and information extraction.
- Text classification and instruction-following workflows.
- Tool calling for integration with external functions.
- Multilingual text processing.
- Open-weight deployment and customization under Apache 2.0.
What improved
- Not yet confirmed by public reporting.
Known flaws & limitations
- Text-only input and output; it does not provide native image, audio or video understanding.
- Generated answers can contain factual errors or fabricated information.
- Performance varies by task, language and deployment conditions; application-specific evaluation remains necessary.
Risks, safety & security issues
- Incorrect or fabricated answers can mislead users when treated as authoritative.
- Generated text can reproduce biases present in training data.
- Harmful or inappropriate responses remain possible; deployments require application-level safeguards.
Good fit for
- Document question answering grounded in retrieved sources.
- Business-text summarization, classification and information extraction.
- Tool-assisted workflows with validated arguments and restricted permissions.
- Locally hosted text applications requiring open weights and deployment control.
Poor fit for
- Native image, audio or video understanding.
- High-stakes decisions without qualified human review.
- Factual answering that requires guaranteed accuracy without source verification.
- Unsupervised execution of consequential actions.
Profile last reviewed 10/21/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.