Granite 4.2 3B vs Granite 4.2 8B
Wins 0 of 6 areas
—
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
Granite 4.2 8B 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 software03Granite 4.2 8B3 of 3 tests
- ReasoningHard problems that need careful thinking02Granite 4.2 8B2 of 3 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own01Granite 4.2 8B1 of 3 tests · 2 ties
- Long documentsFinding answers in very long texts01Granite 4.2 8B1 of 1 test
- Following instructionsDoing exactly what it is asked01Granite 4.2 8B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
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 52% 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 8B pulls ahead
- Reasons across sets of long documentsAA-LCR+20.7points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+7.2points ahead
- Code for real scientific research problemsSciCode+6.2points ahead
Where Granite 4.2 3B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+5.7points ahead
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 8B
- SciCodeGranite 4.2 8B by 6.225.331.5+6.2
- Terminal-Bench 2.1Granite 4.2 8B by 4.513.918.4+4.5
- LiveCodeBench v6Granite 4.2 8B by 3.569.773.2+3.5
AgentsGranite 4.2 8B
- τ-Bench V3 · BankingGranite 4.2 8B by 25.67.6+2
- GDPValtie00tie
- Terminal-Bench 4.0tie00tie
ReasoningGranite 4.2 8B
- GPQA DiamondGranite 4.2 8B by 7.255.963.1+7.2
- Humanity's Last ExamGranite 4.2 8B by 3.16.69.7+3.1
- CritPttie00.3tie
FactsEven
- AA-Omniscience · Non-hallucinationGranite 4.2 3B by 5.773.768+5.7
- AA-Omniscience · AccuracyGranite 4.2 8B by 29.211.2+2
Following instructionsGranite 4.2 8B
- IFBenchGranite 4.2 8B by 574.379.3+5
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 8B by 33.327.861.1+33.3
- Arena-Hard-V2Granite 4.2 8B by 30.23565.2+30.2
- RULER 128KGranite 4.2 8B by 16.155.371.4+16.1
- RULER 64KGranite 4.2 8B by 13.567.581+13.5
- τ³-BenchGranite 4.2 8B by 12.345.858.1+12.3
- HMMT 2025Granite 4.2 8B by 11.666.778.3+11.6
- HMMT Feb25Granite 4.2 8B by 11.666.778.3+11.6
- AIME 2025Granite 4.2 8B by 8.478.386.7+8.4
- AIME25 no toolsGranite 4.2 8B by 8.478.386.7+8.4
- MMLU-ProGranite 4.2 8B by 6.267.874+6.2
- IFBench (prompt)Granite 4.2 8B by 574.379.3+5
- Artificial Analysis Coding IndexGranite 4.2 8B by 4.917.522.4+4.9
- AA-OmniscienceGranite 4.2 3B by 2.5-14.7-17.2+2.5
- BFCLv4Granite 4.2 3B by 2.152.450.3+2.1
- AA IntelligenceGranite 4.2 8B by 29.111.1+2
Questions people ask
Which is better, Granite 4.2 3B or Granite 4.2 8B?
Granite 4.2 8B 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 52% less. They are level on facts.
Which is better for coding?
Granite 4.2 8B. It wins 3 of the 3 coding tests both models report; Granite 4.2 3B wins none.
Which is cheaper?
Granite 4.2 3B costs $0.03 per million input tokens and $0.12 per million output tokens; Granite 4.2 8B costs $0.06 and $0.25. That makes Granite 4.2 3B about 52% 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.