DeepSeek-V3.2 vs MiniMax M2.1
Wins 4 of 7 areas
Reasoning · Facts · Math · Long documents
Wins 3 of 7 areas
Coding · Agents · Following instructions
DeepSeek-V3.2 wins more areas, narrowly.MiniMax M2.1 is better at coding.
Scores updated · 46 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.
- ReasoningHard problems that need careful thinking30DeepSeek-V3.23 of 3 tests
- FactsGetting facts right instead of making them up21DeepSeek-V3.22 of 3 tests
- MathCompetition and research-level math10DeepSeek-V3.21 of 1 test
- Long documentsFinding answers in very long texts10DeepSeek-V3.21 of 1 test
- CodingWriting and fixing software24MiniMax M2.14 of 6 tests
- AgentsCarrying out multi-step tasks on its own02MiniMax M2.12 of 2 tests
- Following instructionsDoing exactly what it is asked01MiniMax M2.11 of 1 test
Math, 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.
DeepSeek-V3.2 costs 53% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where DeepSeek-V3.2 pulls ahead
- Olympiad math problems with short answersIMOAnswerBench+17.9points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+11.8points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+6.8points ahead
Where MiniMax M2.1 pulls ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+21.2points ahead
- Real work tasks from 44 professionsGDPVal+19.9points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+14.2points ahead
Every test, side by side
All 46 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingMiniMax M2.1
- SWE-bench ProMiniMax M2.1 by 21.215.636.8+21.2
- Terminal-Bench HardDeepSeek-V3.2 by 6.835.628.8+6.8
- SWE-bench VerifiedDeepSeek-V3.2 by 6.173.167+6.1
- SWE-bench MultilingualMiniMax M2.1 by 2.370.272.5+2.3
- LMArena · WebDevMiniMax M2.1 by 22 rating points13621384+22 rating
- SciCodeMiniMax M2.1 by 1.838.940.7+1.8
AgentsMiniMax M2.1
- GDPValMiniMax M2.1 by 19.99.829.7+19.9
- BrowseCompMiniMax M2.1 by 10.651.462+10.6
ReasoningDeepSeek-V3.2
- CritPtDeepSeek-V3.2 by 2.62.90.3+2.6
- Humanity's Last ExamDeepSeek-V3.2 by 1.424.623.2+1.4
- GPQA DiamondDeepSeek-V3.2 by 18483+1
FactsDeepSeek-V3.2
- AA-Omniscience · Non-hallucinationMiniMax M2.1 by 14.217.331.5+14.2
- AA-Omniscience · AccuracyDeepSeek-V3.2 by 11.83321.2+11.8
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.2 by 5.56.311.8+5.5
Following instructionsMiniMax M2.1
- IFBenchMiniMax M2.1 by 9.260.769.9+9.2
Other results29 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.
- AA Agentic IndexMiniMax M2.1 by 29.118.347.4+29.1
- HMMT Feb. 2025DeepSeek-V3.2 by 21.592.571+21.5
- Multi-SWE-BenchMiniMax M2.1 by 18.830.649.4+18.8
- BrowseComp-ZHDeepSeek-V3.2 by 17.26547.8+17.2
- HMMT Nov. 2025DeepSeek-V3.2 by 15.79074.3+15.7
- xbench-DeepSearchDeepSeek-V3.2 by 12.755.743+12.7
- xbench-DeepSearch (2025.10)DeepSeek-V3.2 by 12.755.743+12.7
- AIME 2025DeepSeek-V3.2 by 12.193.181+12.1
- Artificial Analysis Coding IndexDeepSeek-V3.2 by 11.444.232.8+11.4
- GAIA (no file)DeepSeek-V3.2 by 10.875.164.3+10.8
- AA-OmniscienceDeepSeek-V3.2 by 10.4-22.5-32.9+10.4
- Terminal-BenchMiniMax M2.1 by 10.237.747.9+10.2
- ToolathlonMiniMax M2.1 by 8.335.243.5+8.3
- SWT-benchMiniMax M2.1 by 7.36269.3+7.3
- τ²-BenchMiniMax M2.1 by 6.780.387+6.7
- AIME25 no toolsDeepSeek-V3.2 by 6.389.383+6.3
- vectara_answer_rateMiniMax M2.1 by 5.992.698.5+5.9
- BrowseComp (context management)DeepSeek-V3.2 by 5.667.662+5.6
- browsecomp_with_context_managerDeepSeek-V3.2 by 5.667.662+5.6
- vectara_factual_consistencyDeepSeek-V3.2 by 5.593.788.2+5.5
- LiveCodeBenchDeepSeek-V3.2 by 5.383.378+5.3
- τ²-Bench Telecom (AA run)DeepSeek-V3.2 by 5.290.685.4+5.2
- vectara_avg_summary_lengthMiniMax M2.1 by 44.9 rating points62106.9+44.9 rating
- ResearchRubricsMiniMax M2.1 by 4.455.860.2+4.4
- MMLU-ProMiniMax M2.1 by 38588+3
- SWE-PerfMiniMax M2.1 by 2.20.93.1+2.2
- Terminal-Bench 2.0MiniMax M2.1 by 1.546.447.9+1.5
- AA Intelligencetie21.520.9tie
- OctoCodingbenchtie2626.1tie
Questions people ask
Which is better, DeepSeek-V3.2 or MiniMax M2.1?
DeepSeek-V3.2 wins four of the seven areas where both have results: reasoning, facts, math and long documents. MiniMax M2.1 wins coding, agents and following instructions.
Which is better for coding?
MiniMax M2.1. It wins 4 of the 6 coding tests both models report; DeepSeek-V3.2 wins 2.
Which is cheaper?
DeepSeek-V3.2 costs $0.28 per million input tokens and $0.42 per million output tokens; MiniMax M2.1 costs $0.30 and $1.20. That makes DeepSeek-V3.2 about 53% cheaper for the same work.
How do you compare the two?
We use the 46 benchmark tests both models have published scores on. The verdict counts the 17 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 29 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.