Across 8 comparable benchmarks, Claude Sonnet 4.6 has better reported results on 8 and worse on 0.
Moderate evidence8 of 8 shared results countedHow we compare
Price basis: Listed rates vs standard rates. The price section below shows the applicable periods.
| Fact | DeepSeek V3.2-Exp | Claude Sonnet 4.6 |
|---|---|---|
| Context window | 128K tokens | 1M tokens |
| Input / 1M | — | $3.00 |
| Output / 1M | — | $15.00 |
| Blended (4:1) | — | $3.46 |
| Status | Available | Available |
| Released | September 29, 2025 | February 17, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 4 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Claude Sonnet 4.6 ahead on 2 of 2 counted
Claude Sonnet 4.6 ahead on 2 of 2 counted
Claude Sonnet 4.6 ahead on 1 of 1 counted
No shared comparable results for Abstract reasoning, Mathematics, Computer & tool use, Long context, Multimodal, Professional work & productivity, Cybersecurity.
Published scores, with matching modes where available. Gaps show DeepSeek V3.2-Exp relative to Claude Sonnet 4.6. Reading the results
| Benchmark | DeepSeek V3.2-Exp | Claude Sonnet 4.6 | Gap |
|---|---|---|---|
| SWE-bench VerifiedRepository Engineering · Accuracy | 67.8Thinking · tools | 79.6≠Thinking · no tools | −11.8 pts#69 vs #20 of 108 |
| Terminal Bench HardAgentic Development · Accuracy | 31.1Thinking · tools | 53.0≠Max · tools | −21.9 pts#66 vs #9 of 157 |
| CritPtScientific Reasoning · Score | 1.4Normal · no tools | 3.1≠Max · no tools | 0.45×#96 vs #78 of 134 |
| GPQA DiamondScientific Reasoning · Accuracy | 79.9Thinking · no tools | 89.9Thinking · no tools | −10.0 pts#113 vs #52 of 200 |
| BrowseCompFact Finding · Accuracy | 40.1Thinking · tools | 74.7Thinking · tools | −34.6 pts#52 vs #33 of 57 |
| HLEKnowledge Exams · Accuracy | 20.3Thinking · tools | 49.0Thinking · tools | −28.7 pts#103 vs #53 of 151 |
| ECICapability Indices · ECI score (capability index, higher is better) | 145Reported best | 152Reported best | 0.95×#79 vs #39 of 168 |
| LiveBenchCross-capability Suites · Accuracy | 58.9Thinking · no tools | 75.3≠Thinking High · no tools | −16.4 pts#88 vs #32 of 106 |
Repository Engineering · Accuracy
Agentic Development · Accuracy
Scientific Reasoning · Score
Scientific Reasoning · Accuracy
Fact Finding · Accuracy
A dash means no result is available.
| Benchmark | DeepSeek V3.2-Exp | Claude Sonnet 4.6 |
|---|---|---|
| AIME2025 | 89.3 | — |
| FrontierMath - Tier 4 | — | 8.3 |
| MMLU-Pro | 85.0 | — |
| SciCode | — | 50.1 |
| Aider-Polyglot | 74.2 | — |
| APEX-SWE | — | 34.5 |
| Code Migration | — | 39.9 |
| DeepSWE | — | 29.9 |
| FrontierCode 1.1 | — | 40.0 |
| LiveCodeBench | 74.1 | — |
| Terminal Bench 2.0 | — | 59.1 |
| Terminal-Bench 1.0 | 37.7 | — |
| Terminal-Bench 4.0 | — | 3.0 |
| Vibe Code Bench v1.1 | — | 51.5 |
| APEX-Agents | — | 43.0 |
| MCP-Atlas | — | 69.5 |
| OSWorld 2.0 | — | 41.5 |
| OSWorld-Verified | — | 72.5 |
| Pinch Bench | — | 88.0 |
| PinchBench v2 | — | 62.7 |
| ExploitBench | — | 25.8 |
| LiveBench 2.0 | — | 65.1 |
| Vals Index | — | 50.6 |
| Creative Writing | — | 1811 |
| Context Arena | — | 82.8 |
| Fiction.liveBench | 83.3 | — |
| GDP.pdf | — | 15.8 |
| GeoBench ACW | — | 73.0 |
| MMMU-Pro | — | 75.6 |
| AA-AnalystAgent | — | 20.0 |
| AutomationBench | — | 5.3 |
| EMB (Excel Modeling) | — | 60.1 |
| Finance Agent v2 | — | 51.0 |
| GDPval-AA | — | 57.0 |
| Harvey Lab-AA | — | 86.0 |
| Harvey's Legal Agent Benchmark | — | 5.0 |
| Legal Research Bench | — | 38.5 |
| Public Benefits Bench v1.1 | — | 62.5 |
| SAGE | — | 46.6 |
| Tax Agent Bench | — | 25.8 |
| τ²-Bench | 66.7 | — |
| τ²-Bench - Telecom | — | 97.9 |
| τ³-Banking | — | 34.4 |
| ARC-AGI-1 | — | 86.5 |
| ARC-AGI-2 | — | 60.4 |
| Vending-Bench 2 | — | 7204 |
| SimpleQA | 97.1 | — |
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.
DeepSeek V3.2-Exp: Listed text rates; pricing tier not confirmed.
Claude Sonnet 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
| Price / 1M tokens | DeepSeek V3.2-Exp | Claude Sonnet 4.6 | Ratio |
|---|---|---|---|
| Input | — | $3.00 | — |
| Output | — | $15.00 | — |
| Cached input | — | $0.30 | — |
| Cache write | — | — | — |
| Blended | — | $3.46 | — |
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 | DeepSeek V3.2-Exp | Claude Sonnet 4.6 |
|---|---|---|
| ARC-AGI-1 | Not reported | |
| ARC-AGI-2 | Not reported | $2.6966 / task Average · ARC-AGI-2 Score: 60.4 Thinking High (With Tools) |
| AutomationBench | Not reported | $0.84 / task Average · Pass Rate: 5.0 Thinking High (With Tools) · zapier-private-heldout 1 more settings$0.77 / task Average · Pass Rate: 3.6 Thinking Medium (With Tools) · zapier-private-heldout |
| DeepSWE | Not reported | |
| FrontierCode 1.1 | Not reported | $2.8979 / task Average · Weighted rubric score: 24.3 Max (With Tools) · frontiercode-1-1-main $2.4401 / task Average · Weighted rubric score: 40.0 Max (With Tools) · frontiercode-1-1-extended $2.2694 / task Average · Weighted rubric score: 39.9 Thinking High (With Tools) · frontiercode-1-1-extended 5 more settings$2.6993 / task Average · Weighted rubric score: 23.5 Thinking High (With Tools) · frontiercode-1-1-main $1.4261 / task Average · Weighted rubric score: 30.4 Thinking Low (With Tools) · frontiercode-1-1-extended $1.6481 / task Average · Weighted rubric score: 15.0 Thinking Low (With Tools) · frontiercode-1-1-main $1.9375 / task Average · Weighted rubric score: 37.5 Thinking Medium (With Tools) · frontiercode-1-1-extended $2.3012 / task Average · Weighted rubric score: 21.1 Thinking Medium (With Tools) · frontiercode-1-1-main |
| LiveBench 2.0 | Not reported | $1.6841 / task Average · Overall score (equal-weight mean of 7 suite scores, 0-100): 65.1 Max (task-specific tools) |
| Tax Agent Bench | Not reported | $1.1935 / task Average · Accuracy (%, rubric × citation quality): 25.8 Unspecified Reasoning (With Tools) |
| Attribute | DeepSeek V3.2-Exp | Claude Sonnet 4.6 |
|---|---|---|
| Availability | Available | Available |
| Provider | DeepSeek-AI | Anthropic |
| Model type | Reasoning model | Chat model |
| First release | September 29, 2025 | February 17, 2026 |
| Context window | 128K tokens | 1M tokens |
| Max output | 64K tokens | 64K tokens |
| Total parameters | 671B | — |
| Active parameters | 37B | — |
| Architecture | Mixture of Experts (MoE) | Undisclosed |
| Documented capabilities | Reasoning | Reasoning, Multilingual |
| Knowledge cutoff | — | 2026-01 |
| Input modalities | Text | Text, Image |
| Output modalities | Text | Text |
| Weights | Open weights | Closed / API only |
| Licence | MIT License | — |
| Commercial use | Commercial use allowed | Proprietary |
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 DeepSeek V3.2-Exp or Claude Sonnet 4.6? 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.