DeepSeek-V3.2 vs Gemma 4 26B A4B
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
Wins 1 of 7 areas
Following instructions
DeepSeek-V3.2 is the stronger all-rounder.Gemma 4 26B A4B is cheaper and better at following instructions.
Scores updated · 43 tests both models report · How we compare
Where each one wins
Tests won in each of the seven 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 software61DeepSeek-V3.26 of 8 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own40DeepSeek-V3.24 of 4 tests
- ReasoningHard problems that need careful thinking30DeepSeek-V3.23 of 3 tests
- MathCompetition and research-level math30DeepSeek-V3.23 of 3 tests
- FactsGetting facts right instead of making them up21DeepSeek-V3.22 of 3 tests
- Long documentsFinding answers in very long texts10DeepSeek-V3.21 of 1 test
- Following instructionsDoing exactly what it is asked01Gemma 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 24% 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 DeepSeek-V3.2 pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+55.7points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+52.9points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+25.1points ahead
Every test, side by side
All 43 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3.2
- SWE-bench VerifiedDeepSeek-V3.2 by 55.773.117.4+55.7
- SWE-bench MultilingualDeepSeek-V3.2 by 52.970.217.3+52.9
- Terminal-Bench HardDeepSeek-V3.2 by 2235.613.6+22
- Terminal-Bench 2.1DeepSeek-V3.2 by 7.846.839+7.8
- LiveCodeBench v6DeepSeek-V3.2 by 6.283.377.1+6.2
- SWE-bench ProDeepSeek-V3.2 by 1.815.613.8+1.8
- SciCodeGemma 4 26B A4B by 1.138.940+1.1
- LMArena · WebDevtie13621359tie
AgentsDeepSeek-V3.2
- BrowseCompDeepSeek-V3.2 by 25.151.426.3+25.1
- MCP AtlasDeepSeek-V3.2 by 12.262.250+12.2
- τ-Bench V3 · BankingDeepSeek-V3.2 by 6.818.812+6.8
- GDPValDeepSeek-V3.2 by 6.49.83.4+6.4
ReasoningDeepSeek-V3.2
- Humanity's Last ExamDeepSeek-V3.2 by 5.324.619.3+5.3
- GPQA DiamondDeepSeek-V3.2 by 4.88479.2+4.8
- CritPtDeepSeek-V3.2 by 2.92.90+2.9
FactsDeepSeek-V3.2
- AA-Omniscience · AccuracyDeepSeek-V3.2 by 13.93319.1+13.9
- AA-Omniscience · Non-hallucinationDeepSeek-V3.2 by 3.717.313.6+3.7
- Vectara HHEM hallucination ratelower is betterGemma 4 26B A4B by 1.16.35.2+1.1
MathDeepSeek-V3.2
- AIME 2026DeepSeek-V3.2 by 5.994.288.3+5.9
- HMMT Feb. 2026DeepSeek-V3.2 by 5.184.179+5.1
- IMOAnswerBenchDeepSeek-V3.2 by 478.374.3+4
Following instructionsGemma 4 26B A4B
- IFBenchGemma 4 26B A4B by 11.760.772.4+11.7
Other results20 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.
- τ²-Bench Telecom (AA run)DeepSeek-V3.2 by 4790.643.6+47
- AA-OmniscienceDeepSeek-V3.2 by 28.3-22.5-50.8+28.3
- MCPMarkDeepSeek-V3.2 by 23.83814.2+23.8
- HLE (with tools)DeepSeek-V3.2 by 23.640.817.2+23.6
- Tool-DecathlonDeepSeek-V3.2 by 23.235.212+23.2
- Terminal-Bench 2.0DeepSeek-V3.2 by 12.246.434.2+12.2
- τ²-BenchDeepSeek-V3.2 by 12.180.368.2+12.1
- τ³-BenchDeepSeek-V3.2 by 10.269.259+10.2
- AA Agentic IndexDeepSeek-V3.2 by 7.318.311+7.3
- vectara_answer_rateGemma 4 26B A4B by 7.292.699.8+7.2
- t2-benchGemma 4 26B A4B by 5.380.285.5+5.3
- vectara_avg_summary_lengthGemma 4 26B A4B by 5.16267.1+5.1
- Artificial Analysis Coding IndexDeepSeek-V3.2 by 4.944.239.3+4.9
- AA IntelligenceDeepSeek-V3.2 by 4.821.516.7+4.8
- HMMT Nov. 2025DeepSeek-V3.2 by 2.59087.5+2.5
- GPQA (unspecified)Gemma 4 26B A4B by 2.479.982.3+2.4
- MMLU-ProDeepSeek-V3.2 by 2.48582.6+2.4
- vectara_factual_consistencyGemma 4 26B A4B by 1.193.794.8+1.1
- MMLU-ReduxDeepSeek-V3.2 by 193.792.7+1
- HMMT Feb. 2025tie92.591.7tie
Questions people ask
Which is better, DeepSeek-V3.2 or Gemma 4 26B A4B?
DeepSeek-V3.2 wins six of the seven areas where both have results: coding, agents, reasoning, facts, math and long documents. Gemma 4 26B A4B wins following instructions, and costs 24% less.
Which is better for coding?
DeepSeek-V3.2. It wins 6 of the 8 coding tests both models report; Gemma 4 26B A4B wins 1, and 1 is a tie.
Which is cheaper?
DeepSeek-V3.2 costs $0.28 per million input tokens and $0.42 per million output tokens; Gemma 4 26B A4B costs $0.13 and $0.40. That makes Gemma 4 26B A4B about 24% cheaper for the same work.
How do you compare the two?
We use the 43 benchmark tests both models have published scores on. The verdict counts the 23 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 20 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.