Claude Opus 4.6 vs DeepSeek-V3.2
Wins 6 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents
Wins 1 of 7 areas
Following instructions
Claude Opus 4.6 is the stronger all-rounder.DeepSeek-V3.2 is cheaper and better at following instructions.
Scores updated · 48 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 software70Claude Opus 4.67 of 7 tests
- ReasoningHard problems that need careful thinking40Claude Opus 4.64 of 4 tests
- AgentsCarrying out multi-step tasks on its own30Claude Opus 4.63 of 4 tests · 1 tie
- FactsGetting facts right instead of making them up31Claude Opus 4.63 of 4 tests
- MathCompetition and research-level math21Claude Opus 4.62 of 3 tests
- Long documentsFinding answers in very long texts10Claude Opus 4.61 of 1 test
- Following instructionsDoing exactly what it is asked01DeepSeek-V3.21 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.
DeepSeek-V3.2 costs 98% 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 Claude Opus 4.6 pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+64.8points ahead
- Real work tasks from 44 professionsGDPVal+46.1points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+37.8points ahead
Where DeepSeek-V3.2 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+7.6points ahead
- Olympiad math problems with short answersIMOAnswerBench+3points ahead
Every test, side by side
All 48 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.6
- SWE-bench ProClaude Opus 4.6 by 37.853.415.6+37.8
- LMArena · WebDevClaude Opus 4.6 by 184 rating points15461362+184 rating
- SciCodeClaude Opus 4.6 by 1351.938.9+13
- Terminal-Bench HardClaude Opus 4.6 by 10.646.235.6+10.6
- SWE-bench VerifiedClaude Opus 4.6 by 7.780.873.1+7.7
- SWE-bench MultilingualClaude Opus 4.6 by 7.677.870.2+7.6
- LiveCodeBench v6Claude Opus 4.6 by 5.588.883.3+5.5
AgentsClaude Opus 4.6
- GDPValClaude Opus 4.6 by 46.155.99.8+46.1
- BrowseCompClaude Opus 4.6 by 32.68451.4+32.6
- AA ApexAgentsClaude Opus 4.6 by 18.53314.5+18.5
- MCP Atlastie62.762.2tie
ReasoningClaude Opus 4.6
- ARC-AGI-2Claude Opus 4.6 by 64.868.84+64.8
- Humanity's Last ExamClaude Opus 4.6 by 15.339.924.6+15.3
- CritPtClaude Opus 4.6 by 9.712.62.9+9.7
- GPQA DiamondClaude Opus 4.6 by 5.689.684+5.6
FactsClaude Opus 4.6
- AA-Omniscience · Non-hallucinationClaude Opus 4.6 by 19.937.217.3+19.9
- SimpleQA VerifiedClaude Opus 4.6 by 19.54727.5+19.5
- AA-Omniscience · AccuracyClaude Opus 4.6 by 144733+14
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.2 by 5.912.26.3+5.9
MathClaude Opus 4.6
- HMMT Feb. 2026Claude Opus 4.6 by 12.196.284.1+12.1
- IMOAnswerBenchDeepSeek-V3.2 by 375.378.3+3
- AIME 2026Claude Opus 4.6 by 2.596.794.2+2.5
Following instructionsDeepSeek-V3.2
- IFBenchDeepSeek-V3.2 by 7.653.160.7+7.6
Other results24 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.
- CyberGymClaude Opus 4.6 by 56.573.817.3+56.5
- AA Agentic IndexClaude Opus 4.6 by 49.367.618.3+49.3
- AA-OmniscienceClaude Opus 4.6 by 36.213.7-22.5+36.2
- ARC-AGI-1Claude Opus 4.6 by 369357+36
- DeepSearchQA (F1)Claude Opus 4.6 by 30.491.360.9+30.4
- ToolathlonClaude Opus 4.6 by 21.656.835.2+21.6
- frontiermath_tier_4_v1Claude Opus 4.6 by 20.822.92.1+20.8
- Terminal-Bench 2.0Claude Opus 4.6 by 1965.446.4+19
- MCPMarkClaude Opus 4.6 by 18.756.738+18.7
- browsecomp_with_context_managerClaude Opus 4.6 by 16.48467.6+16.4
- HLE (with tools)Claude Opus 4.6 by 12.553.340.8+12.5
- Tool-DecathlonClaude Opus 4.6 by 1247.235.2+12
- AA IntelligenceClaude Opus 4.6 by 10.431.921.5+10.4
- vectara_avg_summary_lengthClaude Opus 4.6 by 75.6 rating points137.662+75.6 rating
- vectara_answer_rateClaude Opus 4.6 by 7.299.892.6+7.2
- AIME 2025Claude Opus 4.6 by 6.799.893.1+6.7
- AIME25 no toolsClaude Opus 4.6 by 6.395.689.3+6.3
- HMMT Nov. 2025Claude Opus 4.6 by 6.396.390+6.3
- vectara_factual_consistencyDeepSeek-V3.2 by 5.987.893.7+5.9
- LiveCodeBenchClaude Opus 4.6 by 5.588.883.3+5.5
- MMLU-ProClaude Opus 4.6 by 4.189.185+4.1
- Artificial Analysis Coding IndexClaude Opus 4.6 by 3.948.144.2+3.9
- τ³-BenchClaude Opus 4.6 by 3.272.469.2+3.2
- τ²-Bench Telecom (AA run)Claude Opus 4.6 by 1.592.190.6+1.5
Questions people ask
Which is better, Claude Opus 4.6 or DeepSeek-V3.2?
Claude Opus 4.6 wins six of the seven areas where both have results: coding, agents, reasoning, facts, math and long documents. DeepSeek-V3.2 wins following instructions, and costs 98% less.
Which is better for coding?
Claude Opus 4.6. It wins 7 of the 7 coding tests both models report; DeepSeek-V3.2 wins none.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; DeepSeek-V3.2 costs $0.28 and $0.42. That makes DeepSeek-V3.2 about 98% cheaper for the same work.
How do you compare the two?
We use the 48 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 24 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.