IBM
What its model cards measure
Tall ticks are IBM model cards that list the benchmark in their headline table. Gray dots are cards that leave it out. Rows are the benchmarks the lab listed most, at least twice each, in the order it started using them.
MMLU BIG-Bench Hard GSM8K HumanEval HumanEval+ IFEval AttaQ Arena-Hard-Auto AlpacaEval 2.0 PopQA TruthfulQA DROP MMLU-Pro GPQA Diamond AGIEval GSM-Symbolic MATH DeepMind Mathematics MBPP MBPP+ CRUXEval BigCodeBench Berkeley Function Calling Leaderboard (BFCL) MultiPL-E MMMLU INCLUDE MGSM SALAD-Bench
Model cards
| Date | Model | Document | Benchmarks |
|---|---|---|---|
| 25 Aug 2026 | Granite 4.2 30B | granite-4.2-30b model card (Hugging Face) | 19 |
| 29 Apr 2026 | Granite 4.1 30B | granite-4.1-30b model card (Hugging Face) | 28 |
| 28 Oct 2025 | granite-4.0-1b | granite-4.0-1b model card (Hugging Face) | 23 |
| 2 Oct 2025 | Granite 4.0 H Small | granite-4.0-h-small model card (Hugging Face) | 25 |
| 16 Apr 2025 | Granite 3.3 8B Instruct | granite-3.3-8b-instruct model card (Hugging Face) | 14 |
| 26 Feb 2025 | Granite 3.2 8B Instruct | granite-3.2-8b-instruct model card (Hugging Face) | 12 |
Every IBM model
- Granite 4.2 30B Aug 2026
- Granite 4.1 30B Apr 2026
- Granite 4.1 8B Apr 2026
- Granite 4.0 H Small Oct 2025
- Granite Vision 3.3 2B Jun 2025
- Granite 3.3 8B Instruct Apr 2025
- Granite 3.2 8B Instruct Feb 2025
- Granite Vision 3.2 2B Feb 2025
- Granite-3.1-8B-Instruct Dec 2024
- Granite-3.1-2B-Instruct Dec 2024
- Granite-3.0-8B-Instruct Oct 2024