Inference
EN

Granite 4.2 30B vs Granite 4.2 8B

Change models
Granite 4.2 30BIBM
Granite 4.2 8BIBM
IndexAll categories, out of 100.
41.6 ±5.4 best
37.6 ±5.4
GapIndex points below the top.
Top
−4.0 · no clear gap
CoverageIndex weight that has results.
85%
85%
SpeedOutput tokens per second.
—
101/s
EloLMArena rating and rank.
1361 best
1319
ReleasedMonth of release.
2026-08
2026-08
Input / 1MUS dollars per 1M input tokens.
—
$0.06
Output / 1MUS dollars per 1M output tokens.
—
$0.15
1K chats2K tokens in, 500 out, each.
—
$0.195
1K documents20K tokens in, 1K out, each.
—
$1.35
1K agent tasks200K tokens in, 20K out, each.
—
$15.00
ContextMaximum tokens in one request.
128K
131K best
AgenticMulti-step tasks with tools.
42.0 best
36.8
CodingCode writing and repair.
43.2 best
36.5
ReasoningLogic problems and puzzles.
39.2 best
37.8
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
35.0 best
33.3
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
61.1
63.2 best
MathMath problems.
47.5 best
41.9

Benchmarks

20 shared
Granite 4.2 30B
Granite 4.2 8B
Artificial Analysis Humanity's Last ExamKnowledge
11.2% best138 / 222
9.7%148 / 222
Artificial Analysis Long Context ReasoningReasoning
49.0% best160 / 217
45.0%164 / 217
Critical Physics TasksReasoning
0.3%137 / 214
0.3%137 / 214
Artificial Analysis GPQA DiamondKnowledge
64.4% best145 / 199
63.1%150 / 199
Artificial Analysis Omniscience AccuracyKnowledge
11.2%160 / 164
11.8% best159 / 164
Graduate-Level Google-Proof Q&AKnowledge
66.4% best99 / 141
64.1%106 / 141
GDPval-AA normalizedAgentic
3.7% best97 / 133
0.0%101 / 133
Artificial Analysis SciCodeCoding
37.8% best97 / 114
31.5%104 / 114
Software Engineering Benchmark VerifiedCoding
57.0% best90 / 114
47.7%101 / 114
Terminal-Bench 2.1 · BenchLMAgentic
29.2% best81 / 86
20.6%83 / 86
SWE-bench ProCoding
33.3% best79 / 83
19.1%82 / 83
Massive Multitask Language Understanding ProfessionalKnowledge
77.6% best43 / 75
74.0%53 / 75
LiveCodeBench v6Coding
75.8% best34 / 59
73.2%38 / 59
Instruction Following BenchmarkInstruction
77.2%21 / 57
79.3% best16 / 57
American Invitational Mathematics Examination 2025Math
89.2% best18 / 53
86.7%24 / 53
Scientific Code BenchmarkCoding
38.8% best23 / 35
36.1%27 / 35
τ³-Bench Tool-Agent-User EvaluationAgentic
62.0% best15 / 32
58.1%16 / 32
Berkeley Function Calling Leaderboard v4Agentic
61.4% best13 / 26
52.4%16 / 26
Harvard-MIT Mathematics Tournament February 2025Math
89.2% best11 / 15
78.3%13 / 15
Artificial Analysis Openness IndexKnowledge
72.2%
72.2%