Across 12 comparable benchmarks, Claude Opus 4.6 has better reported results on 11 and worse on 1.
Moderate evidence12 of 12 shared results countedHow we compare
Price basis: Standard rates vs listed rates. The price section below shows the applicable periods.
| Fact | Claude Opus 4.6 | DeepSeek V3.2-Exp |
|---|---|---|
| Context window | 1M tokens | 128K tokens |
| Input / 1M | $5.00 | — |
| Output / 1M | $25.00 | — |
| Blended (4:1) | $5.76 | — |
| Status | Available | Available |
| Released | February 5, 2026 | September 29, 2025 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 6 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Claude Opus 4.6 ahead on 3 of 3 counted
Claude Opus 4.6 ahead on 1 of 1 counted
Claude Opus 4.6 ahead on 2 of 2 counted
Claude Opus 4.6 ahead on 1 of 1 counted
Claude Opus 4.6 ahead on 1 of 1 counted
No shared comparable results for Abstract reasoning, Computer & tool use, Long context, Multimodal, Cybersecurity.
Published scores, with matching modes where available. Gaps show Claude Opus 4.6 relative to DeepSeek V3.2-Exp. Reading the results
| Benchmark | Claude Opus 4.6 | DeepSeek V3.2-Exp | Gap |
|---|---|---|---|
| LiveCodeBenchAlgorithmic Coding · Pass @K | 76.0Extended · no tools | 74.1≠Thinking · no tools | 1.03×#42 vs #47 of 113 |
| SWE-bench VerifiedRepository Engineering · Accuracy | 80.8Extended · tools | 67.8≠Thinking · tools | +13.0 pts#11 vs #69 of 108 |
| Terminal Bench HardAgentic Development · Accuracy | 48.5Normal · tools | 31.1≠Thinking · tools | +17.4 pts#14 vs #66 of 157 |
| AIME2025Mathematics · Accuracy | 99.8Extended · no tools | 89.3≠Thinking · no tools | +10.5 pts#7 vs #35 of 86 |
| CritPtScientific Reasoning · Score | 12.6Max · no tools | 1.4≠Normal · no tools | 9.00×#48 vs #96 of 134 |
| GPQA DiamondScientific Reasoning · Accuracy | 91.3Extended · no tools | 79.9≠Thinking · no tools | +11.4 pts#43 vs #113 of 200 |
| BrowseCompFact Finding · Accuracy | 84.0Thinking · tools + web | 40.1≠Thinking · tools | +43.9 pts#17 vs #52 of 57 |
| τ²-BenchService Workflows · Accuracy | 91.9Extended · tools | 66.7≠Thinking · tools | +25.2 pts#1 vs #26 of 40 |
| Reported separately from the capability domains; not included in the averages. Read each metric’s direction — lower can be better. | |||
| SimpleQAHonesty & Factuality · Accuracy | 72.0Extended · no tools | 97.1≠Thinking · tools | −25.1 pts#7 vs #1 of 46 |
| HLEKnowledge Exams · Accuracy | 53.0Extended · tools + web | 20.3≠Thinking · tools | +32.7 pts#37 vs #103 of 151 |
| ECICapability Indices · ECI score (capability index, higher is better) | 155Reported best | 145Reported best | 1.07×#27 vs #79 of 168 |
| LiveBenchCross-capability Suites · Accuracy | 74.5Thinking High · no tools | 58.9≠Thinking · no tools | +15.6 pts#37 vs #88 of 106 |
Algorithmic Coding · Pass @K
Repository Engineering · Accuracy
Agentic Development · Accuracy
Mathematics · Accuracy
Scientific Reasoning · Score
Scientific Reasoning · Accuracy
Fact Finding · Accuracy
Service Workflows · Accuracy
Honesty & Factuality · Accuracy
A dash means no result is available.
| Benchmark | Claude Opus 4.6 | DeepSeek V3.2-Exp |
|---|---|---|
| AIME 2026 | 96.7 | — |
| FrontierMath | 40.7 | — |
| FrontierMath - Tier 4 | 22.9 | — |
| FrontierMath Tier 4 v2 | 26.8 | — |
| FrontierMath v2 | 66.0 | — |
| HLE-Verified | 46.8 | — |
| HMMT Feb 2026 | 96.2 | — |
| MATH-500 | 97.6 | — |
| MathArena Apex | 34.5 | — |
| MMLU-Pro | — | 85.0 |
| Riemann-bench | 27.2 | — |
| Aider-Polyglot | — | 74.2 |
| APEX-SWE | 41.8 | — |
| FrontierCode 1.1 | 43.7 | — |
| FrontierSWE v1 | 48.9 | — |
| GSO | 41.2 | — |
| HumanEval | 95.0 | — |
| SWE-bench Multilingual | 77.8 | — |
| SWE-Bench Pro - Commercial | 47.1 | — |
| SWE-Bench Pro - Public | 51.9 | — |
| Terminal Bench 2.0 | 65.4 | — |
| Terminal-Bench 1.0 | — | 37.7 |
| Vibe Code Bench v1.1 | 57.6 | — |
| WeirdML v2 | 78.0 | — |
| APEX-Agents | 46.3 | — |
| MCP-Atlas | 76.8 | — |
| OSWorld-Verified | 72.7 | — |
| Pinch Bench | 87.4 | — |
| PinchBench v2 | 69.9 | — |
| Text Arena (Coding) | 1555 | — |
| Creative Writing | 1810 | — |
| Context Arena | 85.2 | — |
| EBR-bench | 12.7 | — |
| Fiction.liveBench | — | 83.3 |
| MMMU-Pro | 77.3 | — |
| SAGE | 51.6 | — |
| τ²-Bench - Telecom | 99.3 | — |
| τ³-Banking | 27.3 | — |
| ARC-AGI-1 | 94.0 | — |
| ARC-AGI-2 | 69.2 | — |
| ARC-AGI-3 | 0.51 | — |
| SimpleBench | 67.6 | — |
| Vending-Bench 2 | 8018 | — |
| MedCode | 49.1 | — |
| MedScribe | 86.7 | — |
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix · 5% steps
Blended price = effective input rate at 90% cache hits × 80% + output rate × 20%
A blended cost needs a published input and output price on both sides, and one of these two does not have them yet.
Claude Opus 4.6: Standard pricing. Anthropic official pricing (platform.claude.com/docs/en/about-claude/pricing) verified 2026-10-21; full 1M context at standard price
DeepSeek V3.2-Exp: Listed text rates; pricing tier not confirmed.
| Price / 1M tokens | Claude Opus 4.6 | DeepSeek V3.2-Exp | Ratio |
|---|---|---|---|
| Input | $5.00 | — | — |
| Output | $25.00 | — | — |
| Cached input | $0.50 | — | — |
| Cache write | — | — | — |
| Blended | $5.76 | — | — |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
| Benchmark | Claude Opus 4.6 | DeepSeek V3.2-Exp |
|---|---|---|
| AIME 2026 | Not reported | |
| ARC-AGI-1 | Not reported | |
| ARC-AGI-2 | $2.7209 / task Average · ARC-AGI-2 Score: 66.3 Thinking Medium (No Tools) | Not reported |
| FrontierCode 1.1 | $3.1206 / task Average · Weighted rubric score: 42.9 Max (With Tools) · frontiercode-1-1-extended $3.7346 / task Average · Weighted rubric score: 26.9 Max (With Tools) · frontiercode-1-1-main $3.9796 / task Average · Weighted rubric score: 26.6 Thinking High (With Tools) · frontiercode-1-1-main 5 more settings$3.2978 / task Average · Weighted rubric score: 43.7 Thinking High (With Tools) · frontiercode-1-1-extended $1.8406 / task Average · Weighted rubric score: 32.9 Thinking Low (With Tools) · frontiercode-1-1-extended $2.101 / task Average · Weighted rubric score: 18.3 Thinking Low (With Tools) · frontiercode-1-1-main $2.3912 / task Average · Weighted rubric score: 38.6 Thinking Medium (With Tools) · frontiercode-1-1-extended $2.7941 / task Average · Weighted rubric score: 23.7 Thinking Medium (With Tools) · frontiercode-1-1-main | Not reported |
| Attribute | Claude Opus 4.6 | DeepSeek V3.2-Exp |
|---|---|---|
| Availability | Available | Available |
| Provider | Anthropic | DeepSeek-AI |
| Model type | Reasoning model | Reasoning model |
| First release | February 5, 2026 | September 29, 2025 |
| Context window | 1M tokens | 128K tokens |
| Max output | 64K tokens | 64K tokens |
| Total parameters | — | 671B |
| Active parameters | — | 37B |
| Architecture | Undisclosed | Mixture of Experts (MoE) |
| Documented capabilities | Reasoning, Multilingual | Reasoning |
| Knowledge cutoff | 2025-05 | — |
| Input modalities | Text, Image | Text |
| Output modalities | Text | Text |
| Weights | Closed / API only | Open weights |
| Licence | — | MIT License |
| Commercial use | Proprietary | Commercial use allowed |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used Claude Opus 4.6 or DeepSeek V3.2-Exp? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.