Inference
EN

Granite 4.2 30B vs LFM3-1.7B

Change models
Granite 4.2 30BIBM
LFM3-1.7BLiquidAI
IndexAll categories, out of 100.
41.6 ±5.4 best
41.5 ±6.5
GapIndex points below the top.
Top
−0.1 · no clear gap
CoverageIndex weight that has results.
85% best
75%
SpeedOutput tokens per second.
—
218/s
EloLMArena rating and rank.
1361
—
ReleasedMonth of release.
2026-08
2026-10
Input / 1MUS dollars per 1M input tokens.
—
$0.08
Output / 1MUS dollars per 1M output tokens.
—
$0.20
1K chats2K tokens in, 500 out, each.
—
$0.26
1K documents20K tokens in, 1K out, each.
—
$1.80
1K agent tasks200K tokens in, 20K out, each.
—
$20.00
ContextMaximum tokens in one request.
128K
262K best
AgenticMulti-step tasks with tools.
42.0
46.4 best
CodingCode writing and repair.
43.2 best
37.0
ReasoningLogic problems and puzzles.
39.2
43.5 best
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
35.0 best
28.8
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
61.1 best
57.6
MathMath problems.
47.5
—

Benchmarks

10 shared
Only Granite 4.2 30B: 10Only LFM3-1.7B: 15
Granite 4.2 30B
LFM3-1.7B
Artificial Analysis Long Context ReasoningReasoning
49.0%160 / 217
52.8% best150 / 217
Artificial Analysis Omniscience AccuracyKnowledge
11.2% best160 / 164
7.1%164 / 164
Graduate-Level Google-Proof Q&AKnowledge
66.4% best99 / 141
63.4%107 / 141
Software Engineering Benchmark VerifiedCoding
57.0% best90 / 114
43.4%106 / 114
Terminal-Bench 2.1 · BenchLMAgentic
29.2% best81 / 86
12.9%85 / 86
LiveCodeBench v6Coding
75.8% best34 / 59
73.5%37 / 59
Instruction Following BenchmarkInstruction
77.2% best21 / 57
73.0%36 / 57
Scientific Code BenchmarkCoding
38.8% best23 / 35
28.5%32 / 35
τ³-Bench Tool-Agent-User EvaluationAgentic
62.0%15 / 32
68.0% best9 / 32
Berkeley Function Calling Leaderboard v4Agentic
61.4%13 / 26
71.4% best8 / 26