Gemma 4 26B A4B vs Qwen2.5 Instruct 72B
Wins 5 of 5 areas
Coding · Reasoning · Facts · Long documents · Following instructions
Wins 0 of 5 areas
—
Gemma 4 26B A4B is the stronger all-rounder.
Scores updated · 27 tests both models report · How we compare
Where each one wins
Tests won in each of the five 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.
- ReasoningHard problems that need careful thinking20Gemma 4 26B A4B2 of 3 tests · 1 tie
- CodingWriting and fixing software21Gemma 4 26B A4B2 of 3 tests
- FactsGetting facts right instead of making them up10Gemma 4 26B A4B1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts10Gemma 4 26B A4B1 of 1 test
- Following instructionsDoing exactly what it is asked10Gemma 4 26B A4B1 of 1 test
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.
Gemma 4 26B A4B costs 45% 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 Gemma 4 26B A4B pulls ahead
- Reasons across sets of long documentsAA-LCR+45.7points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+35.5points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+30.1points ahead
Where Qwen2.5 Instruct 72B pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+6.4points ahead
Every test, side by side
All 27 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemma 4 26B A4B
- SciCodeGemma 4 26B A4B by 13.34026.7+13.3
- Terminal-Bench HardGemma 4 26B A4B by 9.113.64.5+9.1
- SWE-bench VerifiedQwen2.5 Instruct 72B by 6.417.423.8+6.4
ReasoningGemma 4 26B A4B
- GPQA DiamondGemma 4 26B A4B by 30.179.249.1+30.1
- Humanity's Last ExamGemma 4 26B A4B by 15.719.33.6+15.7
- CritPttie00tie
FactsGemma 4 26B A4B
- AA-Omniscience · AccuracyGemma 4 26B A4B by 1.619.117.5+1.6
- AA-Omniscience · Non-hallucinationtie13.614.5tie
Following instructionsGemma 4 26B A4B
- IFBenchGemma 4 26B A4B by 35.572.436.9+35.5
Other results17 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.
- OpenBookQAQwen2.5 Instruct 72B by 47.448.896.2+47.4
- HumanEvalQwen2.5 Instruct 72B by 36.65086.6+36.6
- Artificial Analysis Coding IndexGemma 4 26B A4B by 27.439.311.9+27.4
- MBPPQwen2.5 Instruct 72B by 19.868.488.2+19.8
- MMLU-ProGemma 4 26B A4B by 11.582.671.1+11.5
- MMMLUGemma 4 26B A4B by 11.586.374.8+11.5
- τ²-Bench Telecom (AA run)Gemma 4 26B A4B by 9.143.634.5+9.1
- AA IntelligenceGemma 4 26B A4B by 916.77.7+9
- IFEvalGemma 4 26B A4B by 7.391.484.1+7.3
- C-EvalQwen2.5 Instruct 72B by 6.782.589.2+6.7
- MMLU-ReduxGemma 4 26B A4B by 5.992.786.8+5.9
- WinoGrandeQwen2.5 Instruct 72B by 3.279.182.3+3.2
- AA-OmniscienceGemma 4 26B A4B by 2.3-50.8-53.1+2.3
- ARC-ChallengeQwen2.5 Instruct 72B by 1.792.894.5+1.7
- MMLUtie77.877tie
- HellaSwagtie85.384.8tie
- PIQA (Acc.)tie82.282.6tie
Questions people ask
Which is better, Gemma 4 26B A4B or Qwen2.5 Instruct 72B?
Gemma 4 26B A4B wins all five areas where both have results: coding, reasoning, facts, long documents and following instructions. Qwen2.5 Instruct 72B wins none.
Which is better for coding?
Gemma 4 26B A4B. It wins 2 of the 3 coding tests both models report; Qwen2.5 Instruct 72B wins 1.
Which is cheaper?
Gemma 4 26B A4B costs $0.13 per million input tokens and $0.40 per million output tokens; Qwen2.5 Instruct 72B costs $0.47 and $0.49. That makes Gemma 4 26B A4B about 45% cheaper for the same work.
How do you compare the two?
We use the 27 benchmark tests both models have published scores on. The verdict counts the 10 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 17 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.