Granite 4.2 30B vs Granite 4.2 3B
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
Wins 0 of 6 areas
—
Granite 4.2 30B is the stronger all-rounder.Granite 4.2 3B is cheaper.
Scores updated · 28 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software30Granite 4.2 30B3 of 3 tests
- AgentsCarrying out multi-step tasks on its own20Granite 4.2 30B2 of 3 tests · 1 tie
- ReasoningHard problems that need careful thinking20Granite 4.2 30B2 of 3 tests · 1 tie
- Long documentsFinding answers in very long texts10Granite 4.2 30B1 of 1 test
- Following instructionsDoing exactly what it is asked10Granite 4.2 30B1 of 1 test
- FactsGetting facts right instead of making them up00Even0 each · 2 ties
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Granite 4.2 3B costs 81% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Granite 4.2 30B pulls ahead
- Reasons across sets of long documentsAA-LCR+24.7points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+12.7points ahead
- Code for real scientific research problemsSciCode+12.5points ahead
Where Granite 4.2 3B pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 28 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGranite 4.2 30B
- Terminal-Bench 2.1Granite 4.2 30B by 12.726.613.9+12.7
- SciCodeGranite 4.2 30B by 12.537.825.3+12.5
- LiveCodeBench v6Granite 4.2 30B by 6.175.869.7+6.1
AgentsGranite 4.2 30B
- τ-Bench V3 · BankingGranite 4.2 30B by 8.814.45.6+8.8
- GDPValGranite 4.2 30B by 3.83.80+3.8
- Terminal-Bench 4.0tie00tie
ReasoningGranite 4.2 30B
- GPQA DiamondGranite 4.2 30B by 8.564.455.9+8.5
- Humanity's Last ExamGranite 4.2 30B by 4.611.26.6+4.6
- CritPttie0.30tie
FactsEven
- AA-Omniscience · Accuracytie10.19.2tie
- AA-Omniscience · Non-hallucinationtie74.473.7tie
Following instructionsGranite 4.2 30B
- IFBenchGranite 4.2 30B by 2.977.274.3+2.9
Other results15 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- MMLU-ProX lite (IBM)Granite 4.2 30B by 38.866.627.8+38.8
- Arena-Hard-V2Granite 4.2 30B by 32.967.935+32.9
- RULER 128KGranite 4.2 30B by 26.181.455.3+26.1
- HMMT 2025Granite 4.2 30B by 22.589.266.7+22.5
- HMMT Feb25Granite 4.2 30B by 22.589.266.7+22.5
- RULER 64KGranite 4.2 30B by 22.59067.5+22.5
- τ³-BenchGranite 4.2 30B by 16.26245.8+16.2
- Artificial Analysis Coding IndexGranite 4.2 30B by 12.429.917.5+12.4
- AIME 2025Granite 4.2 30B by 10.989.278.3+10.9
- AIME25 no toolsGranite 4.2 30B by 10.989.278.3+10.9
- MMLU-ProGranite 4.2 30B by 9.877.667.8+9.8
- BFCLv4Granite 4.2 30B by 961.452.4+9
- AA IntelligenceGranite 4.2 30B by 3.612.79.1+3.6
- IFBench (prompt)Granite 4.2 30B by 2.977.274.3+2.9
- AA-OmniscienceGranite 4.2 30B by 1.8-12.9-14.7+1.8
Questions people ask
Which is better, Granite 4.2 30B or Granite 4.2 3B?
Granite 4.2 30B wins five of the six areas where both have results: coding, agents, reasoning, long documents and following instructions. Granite 4.2 3B wins none, but costs 81% less. They are level on facts.
Which is better for coding?
Granite 4.2 30B. It wins 3 of the 3 coding tests both models report; Granite 4.2 3B wins none.
Which is cheaper?
Granite 4.2 30B costs $0.16 per million input tokens and $0.65 per million output tokens; Granite 4.2 3B costs $0.03 and $0.12. That makes Granite 4.2 3B about 81% cheaper for the same work.
How do you compare the two?
We use the 28 benchmark tests both models have published scores on. The verdict counts the 13 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 15 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.