Model comparison
Claude Opus 5 vs DeepSeek V3.2
Head-to-head evidence from 9 shared benchmark results across 3 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: Claude Opus 5 #1 (Estimated); DeepSeek V3.2 #88 (Supported). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. Claude Opus 5 and DeepSeek V3.2 share 9 comparable benchmark results. 1 of 8 categories are comparable. 70 results are unique to Claude Opus 5; 10 to DeepSeek V3.2.
Updated July 24, 2026- Shared results
- 9
- Claude Opus 5 only
- 70
- DeepSeek V3.2 only
- 10
- Comparable categories
- 1 / 8
Pick Claude Opus 5 if you want the stronger benchmark profile. DeepSeek V3.2 only becomes the better choice if you want the cheaper token bill or you would rather avoid the extra latency and token burn of a reasoning model.
Confidence note. This is a partial-evidence comparison with 9 shared benchmark results across 3 evidence categories; 1 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
Claude Opus 5 is clearly ahead on the BenchAlign aggregate, 85.88 to 54.55. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Claude Opus 5's sharpest advantage is in coding, where it averages 89.5 against 60.9.
Claude Opus 5 is also the more expensive model on tokens at $5.00 input / $25.00 output per 1M tokens, versus $0.28 input / $0.42 output per 1M tokens for DeepSeek V3.2. That is roughly 59.5x on output cost alone. Claude Opus 5 is the reasoning model in the pair, while DeepSeek V3.2 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | Claude Opus 5 | Δ | DeepSeek V3.2 |
|---|---|---|---|
| Coding | Claude Opus 589.5 | Margin← 28.6 | DeepSeek V3.260.9 |
| Agentic | Claude Opus 590.8 | MarginNo overlap | DeepSeek V3.2Not measured |
| Reasoning | Claude Opus 590.4 | MarginNo overlap | DeepSeek V3.2Not measured |
| Knowledge | Claude Opus 564.7 | MarginNo overlap | DeepSeek V3.2Not measured |
| Math | Claude Opus 5Not measured | MarginNo overlap | DeepSeek V3.217.1 |
| Multimodal | Claude Opus 566.9 | MarginNo overlap | DeepSeek V3.2Not measured |
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | Claude Opus 5 | DeepSeek V3.2 | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | Claude Opus 5$5 input / $25 output | DeepSeek V3.2$0.28 input / $0.42 output | DeepSeek V3.2 has the lower combined listed price. |
| Generation speedtokens per second | Claude Opus 5Not available | DeepSeek V3.235 tok/s | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | Claude Opus 5Not available | DeepSeek V3.23.75 s | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | Claude Opus 5Not available | DeepSeek V3.2128K | A complete context comparison is not available. |
Benchmark Deep Dive
Agentic28 benchmarks
| Benchmark | Claude Opus 5 | DeepSeek V3.2 | Result |
|---|---|---|---|
| FrontierBench v0.1Source | 43.3% | — | Not comparable |
| BrowseCompSource | 90.8% | — | Not comparable |
| HLE w/ toolsSource | 64.7% | — | Not comparable |
| DeepSearchQASource | 95.0% | — | Not comparable |
| DRACOSource | 88.6% | — | Not comparable |
| BrowseComp (10-agent, prerelease)Source | 93.6% | — | Not comparable |
| OSWorld 2.0Source | 70.6% | — | Not comparable |
| MCP AtlasSource | 85.8% | — | Not comparable |
| MCP-Atlas claim coverageSource | 89.1% | — | Not comparable |
| LAB all-pass (Anthropic harness)Source | 23.58% | — | Not comparable |
| LAB criterion-pass (Anthropic harness)Source | 93.74% | — | Not comparable |
| LAB all-pass (Harvey held-out)Source | 11.7% | — | Not comparable |
| LAB criterion-pass (Harvey held-out)Source | 94.1% | — | Not comparable |
| GDPval-AASource | 1861 | — | Not comparable |
| Toolathlon-VerifiedSource | 80.6% | — | Not comparable |
| Toolathlon Verified Pass@3Source | 87.0% | — | Not comparable |
| Toolathlon Verified Pass³Source | 73.1% | — | Not comparable |
| Toolathlon Verified avg. turnsSource | 23.5 turns | — | Not comparable |
| AutomationBenchSource | 26.0% | — | Not comparable |
| AA Agentic IndexSource | 55.3% | — | Not comparable |
| GDPval-AASource | 68.0% | — | Not comparable |
| AA Tau3 BankingSource | 30.3% | — | Not comparable |
| AA BriefcaseSource | 1720 | — | Not comparable |
| aaTerminalBench21Source | 89.1% | — | Not comparable |
| Claw-EvalSource | — | 40.2% | Not comparable |
| VITA-BenchSource | — | 18.5% | Not comparable |
| τ²-bench resultsSource | — | 78.9% | Not comparable |
| Gert LabsSource | — | 29.57% | Not comparable |
CodingClaude Opus 5 wins14 benchmarks
| Benchmark | Claude Opus 5 | DeepSeek V3.2 | Result |
|---|---|---|---|
| SWE-bench VerifiedSource | 96% | — | Not comparable |
| SWE-bench ProSource | 79.2% | — | Not comparable |
| SWE MultilingualSource | 89.5% | — | Not comparable |
| SWE MultimodalSource | 59.4% | — | Not comparable |
| deepSweSource | 68.8% | — | Not comparable |
| FrontierCode 1.1 MainSource | 53.4% | — | Not comparable |
| FrontierCode 1.1 ExtendedSource | 63.6% | — | Not comparable |
| ProgramBench (episode 1)Source | 83.0% | — | Not comparable |
| ProgramBenchSource | 93.0% | — | Not comparable |
| cursorBench32Source | 70.0% | — | Not comparable |
| AA Coding IndexSource | 78.0% | — | Not comparable |
| AA-SciCodeSource | 55.7% | 38.7% | Claude Opus 5 leads |
| SWE-RebenchSource | — | 60.9% | Not comparable |
| React Native EvalsSource | — | 71.5% | Not comparable |
Reasoning5 benchmarks
Knowledge21 benchmarks
| Benchmark | Claude Opus 5 | DeepSeek V3.2 | Result |
|---|---|---|---|
| HLESource | 64.7% | — | Not comparable |
| HLE w/o toolsSource | 56.3% | — | Not comparable |
| HealthBench (raw)Source | 67.1% | — | Not comparable |
| HealthBench (length-adjusted)Source | 57.8% | — | Not comparable |
| HealthBench ProfessionalSource | 59.8% | — | Not comparable |
| HealthBench Professional (raw)Source | 73.4% | — | Not comparable |
| BioMysteryBench (human-solvable)Source | 90.1% | — | Not comparable |
| BioMysteryBench (human-difficult)Source | 49.4% | — | Not comparable |
| SpatialBench VerifiedSource | 72.5% | — | Not comparable |
| SingleCellBenchSource | 60.6% | — | Not comparable |
| ProteinGym HardSource | 47.7% | — | Not comparable |
| Protein DesignSource | 42.5% | — | Not comparable |
| Organic chemistry V2Source | 61.6% | — | Not comparable |
| Protocols (troubleshooting)Source | 61.1% | — | Not comparable |
| Protocols (understanding)Source | 78.4% | — | Not comparable |
| Artificial Analysis Intelligence IndexSource | 60.7% | 24.7% | Claude Opus 5 leads |
| AA-GPQA DiamondSource | 93.2% | 75.1% | Claude Opus 5 leads |
| AA-HLESource | 52.6% | 10.5% | Claude Opus 5 leads |
| AA-Omniscience IndexSource | 31.3% | -46.7% | Claude Opus 5 leads |
| AA-Omniscience AccuracySource | 54.2% | 24.2% | Claude Opus 5 leads |
| AA-Omniscience Hallucination RateSource | 50.1% | 93.5% | Claude Opus 5 leads |
Math7 benchmarks
| Benchmark | Claude Opus 5 | DeepSeek V3.2 | Result |
|---|---|---|---|
| IMO 2026Source | 42/42 | — | Not comparable |
| RiemannBench (no tools)Source | 60.0% | — | Not comparable |
| RiemannBench (tools)Source | 79.0% | — | Not comparable |
| ArXivMath Jun. 2026 (no tools)Source | 90.8% | — | Not comparable |
| ArXivMath Jun. 2026 (tools)Source | 91.3% | — | Not comparable |
| FrontierMath v2 (Tiers 1-3)Source | — | 22.100% | Not comparable |
| FrontierMath v2 (Tier 4)Source | — | 2.100% | Not comparable |
Multilingual3 benchmarks
Multimodal10 benchmarks
| Benchmark | Claude Opus 5 | DeepSeek V3.2 | Result |
|---|---|---|---|
| Chartography (no tools)Source | 29.6% | — | Not comparable |
| Chartography (tools)Source | 83.0% | — | Not comparable |
| BenchCAD Vision2Code (no tools)Source | 0.366 | — | Not comparable |
| BenchCAD Vision2Code (tools)Source | 0.821 | — | Not comparable |
| GDP.pdf (no tools)Source | 83.4% | — | Not comparable |
| GDP.pdf (tools)Source | 85.5% | — | Not comparable |
| OfficeQASource | 78.1% | — | Not comparable |
| OfficeQA ProSource | 66.9% | — | Not comparable |
| AA-MMMU-ProSource | 84.7% | — | Not comparable |
| Design Arena WebsiteSource | — | 1204 | Not comparable |
Inst. Following1 benchmarks
| Benchmark | Claude Opus 5 | DeepSeek V3.2 | Result |
|---|---|---|---|
| AA-IFBenchSource | — | 49.0% | Not comparable |
Frequently Asked Questions (2)
Which is better, Claude Opus 5 or DeepSeek V3.2?
Claude Opus 5 is ahead on BenchLM's BenchAlign leaderboard, 85.88 to 54.55.
Which is better for coding, Claude Opus 5 or DeepSeek V3.2?
Claude Opus 5 has the edge for coding in this comparison, averaging 89.5 versus 60.9. Inside this category, AA-SciCode is the benchmark that creates the most daylight between them.