Inference
EN

Granite-4.0-H-1B vs Llama 3.1 8B

Change models
Granite-4.0-H-1BIBM
Llama 3.1 8BMeta
IndexAll categories, out of 100.
16.3 ±6.9
17.9 ±9.6 best
GapIndex points below the top.
−1.6 · no clear gap
Top
CoverageIndex weight that has results.
80% best
55%
SpeedOutput tokens per second.
—
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2025-10
2024-07
Input / 1MUS dollars per 1M input tokens.
—
—
Output / 1MUS dollars per 1M output tokens.
—
—
1K chats2K tokens in, 500 out, each.
—
—
1K documents20K tokens in, 1K out, each.
—
—
1K agent tasks200K tokens in, 20K out, each.
—
—
ContextMaximum tokens in one request.
128K
128K
AgenticMulti-step tasks with tools.
15.8
—
CodingCode writing and repair.
14.3 best
13.9
ReasoningLogic problems and puzzles.
24.5
—
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
11.1
14.4 best
MultilingualTasks in many languages.
5.0
24.8 best
InstructionTasks with strict prompt rules.
17.4
25.1 best
MathMath problems.
—
10.0

Benchmarks

8 shared
Only Granite-4.0-H-1B: 5Only Llama 3.1 8B: 12
Granite-4.0-H-1B
Llama 3.1 8B
Artificial Analysis Humanity's Last ExamKnowledge
5.0% best176 / 222
4.3%190 / 222
Artificial Analysis GPQA DiamondKnowledge
26.3%197 / 199
27.0% best196 / 199
Graduate-Level Google-Proof Q&AKnowledge
29.7%137 / 141
30.4% best134 / 141
Massive Multitask Language Understanding ProfessionalKnowledge
32.9%70 / 75
48.3% best67 / 75
Instruction-Following EvalInstruction
78.5%57 / 70
80.4% best54 / 70
Massive Multitask Language UnderstandingKnowledge
59.7%31 / 34
73.0% best29 / 34
Evaluating Large Language Models Trained on CodeCoding
73.0% best18 / 22
72.6%19 / 22
Multilingual Grade School MathMultilingual
27.5%11 / 13
68.9% best8 / 13