DeepSeek-V3.2 vs GLM-4.6
Wins 7 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents · Following instructions
Wins 0 of 7 areas
—
DeepSeek-V3.2 is the stronger all-rounder.
Scores updated · 47 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 software51DeepSeek-V3.25 of 8 tests · 2 ties
- ReasoningHard problems that need careful thinking30DeepSeek-V3.23 of 3 tests
- FactsGetting facts right instead of making them up30DeepSeek-V3.23 of 3 tests
- AgentsCarrying out multi-step tasks on its own21DeepSeek-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
- Following instructionsDoing exactly what it is asked10DeepSeek-V3.21 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 75% 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
- Reasons across sets of long documentsAA-LCR+19.3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+17.3points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+16.4points ahead
Where GLM-4.6 pulls ahead
- Real work tasks from 44 professionsGDPVal+3.2points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+2.6points ahead
Every test, side by side
All 47 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3.2
- SWE-bench MultilingualDeepSeek-V3.2 by 16.470.253.8+16.4
- Terminal-Bench HardDeepSeek-V3.2 by 10.635.625+10.6
- SWE-bench ProDeepSeek-V3.2 by 5.915.69.7+5.9
- SWE-bench VerifiedDeepSeek-V3.2 by 5.173.168+5.1
- Terminal-Bench 2.1GLM-4.6 by 2.646.849.4+2.6
- LMArena · WebDevDeepSeek-V3.2 by 21 rating points13621341+21 rating
- LiveCodeBench v6tie83.382.8tie
- SciCodetie38.938.4tie
AgentsDeepSeek-V3.2
- BrowseCompDeepSeek-V3.2 by 6.351.445.1+6.3
- τ-Bench V3 · BankingDeepSeek-V3.2 by 5.418.813.4+5.4
- GDPValGLM-4.6 by 3.29.813+3.2
ReasoningDeepSeek-V3.2
- Humanity's Last ExamDeepSeek-V3.2 by 10.124.614.5+10.1
- GPQA DiamondDeepSeek-V3.2 by 68478+6
- CritPtDeepSeek-V3.2 by 1.82.91.1+1.8
FactsDeepSeek-V3.2
- AA-Omniscience · Non-hallucinationDeepSeek-V3.2 by 11.417.35.9+11.4
- AA-Omniscience · AccuracyDeepSeek-V3.2 by 6.13326.9+6.1
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.2 by 3.26.39.5+3.2
Following instructionsDeepSeek-V3.2
- IFBenchDeepSeek-V3.2 by 17.360.743.4+17.3
Other results27 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.
- Terminal-Bench 2.0DeepSeek-V3.2 by 21.946.424.5+21.9
- τ²-Bench Telecom (AA run)DeepSeek-V3.2 by 20.190.670.5+20.1
- AA-OmniscienceDeepSeek-V3.2 by 19.4-22.5-41.9+19.4
- BrowseComp-ZHDeepSeek-V3.2 by 15.56549.5+15.5
- vectara_avg_summary_lengthGLM-4.6 by 15.26277.2+15.2
- xbench-DeepSearchGLM-4.6 by 14.355.770+14.3
- HLE (with tools)DeepSeek-V3.2 by 10.440.830.4+10.4
- browsecomp_with_context_managerDeepSeek-V3.2 by 10.167.657.5+10.1
- GAIA (text only)GLM-4.6 by 8.463.571.9+8.4
- τ²-BenchDeepSeek-V3.2 by 5.180.375.2+5.1
- ArtifactsBenchGLM-4.6 by 455.859.8+4
- LiveCodeBenchDeepSeek-V3.2 by 3.883.379.5+3.8
- HMMT Feb. 2025DeepSeek-V3.2 by 3.392.589.2+3.3
- vectara_factual_consistencyDeepSeek-V3.2 by 3.293.790.5+3.2
- AA IntelligenceDeepSeek-V3.2 by 321.518.5+3
- FinSearchComp-globalGLM-4.6 by 326.229.2+3
- Terminal-BenchGLM-4.6 by 2.837.740.5+2.8
- vectara_answer_rateGLM-4.6 by 1.992.694.5+1.9
- MMLU-ProDeepSeek-V3.2 by 1.88583.2+1.8
- HMMT Nov. 2025GLM-4.6 by 1.79091.7+1.7
- Artificial Analysis Coding IndexGLM-4.6 by 1.644.245.8+1.6
- HMMT 2025DeepSeek-V3.2 by 1.590.288.7+1.5
- theagentcompanyGLM-4.6 by 13435+1
- AIME 2025tie93.193.9tie
- Multi-SWE-Benchtie30.630tie
- AA Agentic Indextie18.318.6tie
- frontiermath_tier_4_v1tie2.12.1tie
Questions people ask
Which is better, DeepSeek-V3.2 or GLM-4.6?
DeepSeek-V3.2 wins all seven areas where both have results: coding, agents, reasoning, facts, math, long documents and following instructions. GLM-4.6 wins none.
Which is better for coding?
DeepSeek-V3.2. It wins 5 of the 8 coding tests both models report; GLM-4.6 wins 1, and 2 are ties.
Which is cheaper?
DeepSeek-V3.2 costs $0.28 per million input tokens and $0.42 per million output tokens; GLM-4.6 costs $0.57 and $2.20. That makes DeepSeek-V3.2 about 75% cheaper for the same work.
How do you compare the two?
We use the 47 benchmark tests both models have published scores on. The verdict counts the 20 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 27 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.