Benchmark Almanac

Granite 4.0 H Small

Granite 4.0 H Small places in the top three on AttaQ.

IBM Released October 2nd, 2025 Family Granite 4.0 The lab's model card Compare with others
36
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
IBM Oct 2025

Granite 4.0 H Small

Open weights Flagship

118.5 #353 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • SimpleQA Verified 8.57% #50 /56
  • BigCodeBench 46.2% #8 /14
  • GPQA Main 38.3% #49 /60

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
Vectara Hallucination Leaderboard (2025 dataset) 5.20% #13 of 104 Official leaderboard
AGIEval 70.6% #12 of 40 Lab model card
IFEval 87.5% #58 of 135 Lab model card
BIG-Bench Hard 81.6% #33 of 73 Lab model card
MMLU 78.4% #84 of 131 Lab model card
LMArena (Elo) 1,287 #241 of 335 Official leaderboard
INCLUDE 64% #24 of 31 Lab model card
AlpacaEval 2.0 42.5% #13 of 16 Lab model card
MMMLU 69.7% #59 of 71 Lab model card
Wer wird Millionär? LLM Benchmark 149.0 #66 of 77 Official leaderboard
Global MMLU Lite 75% #37 of 43 Official leaderboard
SimpleQA Verified 8.57% #50 of 56 Official leaderboard
Arena-Hard-Auto 46.5% #35 of 39 Lab model card
MMLU-Pro 55.5% #150 of 167 Lab model card
MGSM 38.7% #54 of 57 Lab model card

Math and science

BenchmarkScoreRankSource
GSM-Symbolic 87.4% #1 of 3 Lab model card
MATH 74% #38 of 99 Lab model card
DeepMind Mathematics 59.3% #2 of 4 Lab model card
GSM8K 87.3% #54 of 92 Lab model card
GPQA Main 38.3% #49 of 60 Official leaderboard
GPQA Diamond 40.6% #270 of 302 Lab model card

Coding

BenchmarkScoreRankSource
MBPP 84% #13 of 71 Lab model card
HumanEval 88% #27 of 101 Lab model card
BigCodeBench 46.2% #8 of 14 Lab model card
HumanEval+ 83% #16 of 27 Lab model card
MBPP+ 71% #11 of 15 Lab model card
CRUXEval 50.3% #8 of 9 Lab model card
MultiPL-E 57.4% #17 of 18 Lab model card

Agents and tools

BenchmarkScoreRankSource
Berkeley Function Calling Leaderboard (BFCL) 64.7% #29 of 41 Lab model card

Safety and security

BenchmarkScoreRankSource
Anthropic Red Team 99.8% #6 of 75 Official leaderboard
AIR-Bench 2024 82% #20 of 82 Official leaderboard
SimpleSafetyTests 100% #23 of 75 Official leaderboard
SALAD-Bench 97.3% #1 of 3 Lab model card
AttaQ 86.6% #2 of 5 Lab model card
HELM Safety 91.4 #38 of 75 Official leaderboard

Other

BenchmarkScoreRankSource
HELM Capabilities 57.5 #46 of 60 Official leaderboard

Its lab's numbers put it at 118.5, number 353 of 404

The Self-Reported Index fits one score to the 21 capability results IBM reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 115.7 to 121.3. How we compute it.

Its lab's number next to a leaderboard

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
IFEval 87.5% IFEval Average 89% Leaderboard, strict, share of instructions followed −1.4 pts

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

granite-4.0-h-small model card (Hugging Face) Oct 2, 2025