Claude Sonnet 4.5 vs DeepSeek-V3.2
Wins 2 of 7 areas
Coding · Facts
Wins 3 of 7 areas
Reasoning · Math · Following instructions
DeepSeek-V3.2 wins more areas, narrowly.Claude Sonnet 4.5 is better at coding.
Scores updated · 77 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 thinking12DeepSeek-V3.22 of 4 tests · 1 tie
- MathCompetition and research-level math01DeepSeek-V3.21 of 1 test
- Following instructionsDoing exactly what it is asked01DeepSeek-V3.21 of 1 test
- CodingWriting and fixing software52Claude Sonnet 4.55 of 8 tests · 1 tie
- FactsGetting facts right instead of making them up21Claude Sonnet 4.52 of 4 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own22Even2 each
- Long documentsFinding answers in very long texts11Even1 each
Math 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 96% 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
- Finds hard-to-locate facts by browsing the webBrowseComp+27.3points ahead
- Recent programming contest problemsLiveCodeBench v6+19.3points ahead
- Multi-step tasks using many real software toolsMCP Atlas+18.4points ahead
Where Claude Sonnet 4.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+33.6points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+28points ahead
- Real work tasks from 44 professionsGDPVal+10.7points ahead
Every test, side by side
All 77 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4.5
- SWE-bench ProClaude Sonnet 4.5 by 2843.615.6+28
- LiveCodeBench v6DeepSeek-V3.2 by 19.36483.3+19.3
- Terminal-Bench 2.1Claude Sonnet 4.5 by 955.846.8+9
- SciCodeClaude Sonnet 4.5 by 6.845.738.9+6.8
- SWE-bench VerifiedClaude Sonnet 4.5 by 4.177.273.1+4.1
- SWE-bench MultilingualDeepSeek-V3.2 by 3.26770.2+3.2
- LMArena · WebDevClaude Sonnet 4.5 by 31 rating points13931362+31 rating
- Terminal-Bench Hardtie35.635.6tie
AgentsEven
- BrowseCompDeepSeek-V3.2 by 27.324.151.4+27.3
- MCP AtlasDeepSeek-V3.2 by 18.443.862.2+18.4
- GDPValClaude Sonnet 4.5 by 10.720.59.8+10.7
- τ-Bench V3 · BankingClaude Sonnet 4.5 by 5.724.518.8+5.7
ReasoningDeepSeek-V3.2
- ARC-AGI-2Claude Sonnet 4.5 by 9.613.64+9.6
- Humanity's Last ExamDeepSeek-V3.2 by 6.817.824.6+6.8
- CritPtDeepSeek-V3.2 by 1.81.12.9+1.8
- GPQA Diamondtie83.484tie
FactsClaude Sonnet 4.5
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.5 by 33.650.917.3+33.6
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.2 by 5.7126.3+5.7
- SimpleQA VerifiedClaude Sonnet 4.5 by 3.230.727.5+3.2
- AA-Omniscience · Accuracytie32.933tie
Long documentsEven
- LongBench v2Claude Sonnet 4.5 by 261.859.8+2
- AA-LCRDeepSeek-V3.2 by 172.373.3+1
Following instructionsDeepSeek-V3.2
- IFBenchDeepSeek-V3.2 by 3.457.360.7+3.4
Other results53 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.
- BrowseComp (context management)DeepSeek-V3.2 by 41.526.167.6+41.5
- FinSearchComp-globalClaude Sonnet 4.5 by 34.660.826.2+34.6
- HMMT Feb. 2025DeepSeek-V3.2 by 2567.592.5+25
- BrowseComp-ZHDeepSeek-V3.2 by 22.642.465+22.6
- AA-OmniscienceClaude Sonnet 4.5 by 22.4-0.1-22.5+22.4
- τ²-BenchClaude Sonnet 4.5 by 17.79880.3+17.7
- FinSearchComp-T3 (Tools-allowed)Claude Sonnet 4.5 by 174427+17
- FinSearchComp-T3 w/ toolsClaude Sonnet 4.5 by 174427+17
- BrowseComp w/ toolsDeepSeek-V3.2 by 1624.140.1+16
- HMMT 2025DeepSeek-V3.2 by 15.674.690.2+15.6
- Seal-0 w/ toolsClaude Sonnet 4.5 by 14.953.438.5+14.9
- Multi-SWE-BenchClaude Sonnet 4.5 by 13.744.330.6+13.7
- Terminal-Bench w/ simulated tools (JSON)Claude Sonnet 4.5 by 13.35137.7+13.3
- τ²-Bench Telecom (AA run)DeepSeek-V3.2 by 12.578.190.6+12.5
- LiveCodeBenchDeepSeek-V3.2 by 12.37183.3+12.3
- Terminal-BenchClaude Sonnet 4.5 by 12.35037.7+12.3
- Arena-Hard (Creative Writing)DeepSeek-V3.2 by 12.176.788.8+12.1
- swe_bench_bashClaude Sonnet 4.5 by 11.471.460+11.4
- xbench-DeepSearchClaude Sonnet 4.5 by 10.36655.7+10.3
- LiveCodeBenchV6 no toolsDeepSeek-V3.2 by 10.16474.1+10.1
- Arena-Hard (Hard Prompt)Claude Sonnet 4.5 by 9.963.353.4+9.9
- HMMT25 no toolsDeepSeek-V3.2 by 974.683.6+9
- HLE (with tools)DeepSeek-V3.2 by 8.83240.8+8.8
- HMMT Nov. 2025DeepSeek-V3.2 by 8.381.790+8.3
- Artificial Analysis Coding IndexClaude Sonnet 4.5 by 7.952.144.2+7.9
- OJ-Bench (cpp)DeepSeek-V3.2 by 7.830.438.2+7.8
- OJ-Bench (cpp) no toolsDeepSeek-V3.2 by 7.830.438.2+7.8
- GAIA (text only)Claude Sonnet 4.5 by 7.771.263.5+7.7
- SWT-benchClaude Sonnet 4.5 by 7.569.562+7.5
- Longform Writing eval (Kimi K2 Thinking system card)Claude Sonnet 4.5 by 7.379.872.5+7.3
- theagentcompanyClaude Sonnet 4.5 by 74134+7
- ARC-AGI-1Claude Sonnet 4.5 by 6.763.757+6.7
- vectara_avg_summary_lengthClaude Sonnet 4.5 by 65.8 rating points127.862+65.8 rating
- AIME 2025DeepSeek-V3.2 by 6.18793.1+6.1
- ToolathlonClaude Sonnet 4.5 by 5.84135.2+5.8
- ArtifactsBenchClaude Sonnet 4.5 by 5.761.555.8+5.7
- vectara_factual_consistencyDeepSeek-V3.2 by 5.78893.7+5.7
- BrowseComp-ZH w/ toolsDeepSeek-V3.2 by 5.542.447.9+5.5
- Frames w/ toolsClaude Sonnet 4.5 by 4.88580.2+4.8
- Terminal-Bench 2.0Claude Sonnet 4.5 by 3.65046.4+3.6
- GPQA (unspecified)Claude Sonnet 4.5 by 3.583.479.9+3.5
- OctoCodingbenchDeepSeek-V3.2 by 3.222.826+3.2
- vectara_answer_rateClaude Sonnet 4.5 by 395.692.6+3
- HealthBenchDeepSeek-V3.2 by 2.744.246.9+2.7
- HealthBench no toolsDeepSeek-V3.2 by 2.744.246.9+2.7
- MMLU-ProClaude Sonnet 4.5 by 2.587.585+2.5
- frontiermath_tier_4_v1Claude Sonnet 4.5 by 2.14.22.1+2.1
- SWE-PerfClaude Sonnet 4.5 by 2.130.9+2.1
- MMLU-ReduxClaude Sonnet 4.5 by 1.995.693.7+1.9
- AIME25 no toolsDeepSeek-V3.2 by 1.38889.3+1.3
- AA Agentic Indextie17.518.3tie
- AA Intelligencetie20.721.5tie
- MRCRtie55.455.5tie
Questions people ask
Which is better, Claude Sonnet 4.5 or DeepSeek-V3.2?
DeepSeek-V3.2 wins three of the seven areas where both have results: reasoning, math and following instructions. Claude Sonnet 4.5 wins coding and facts. They are level on agents and long documents.
Which is better for coding?
Claude Sonnet 4.5. It wins 5 of the 8 coding tests both models report; DeepSeek-V3.2 wins 2, and 1 is a tie.
Which is cheaper?
Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; DeepSeek-V3.2 costs $0.28 and $0.42. That makes DeepSeek-V3.2 about 96% cheaper for the same work.
How do you compare the two?
We use the 77 benchmark tests both models have published scores on. The verdict counts the 24 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 53 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.