Model comparison
Claude Sonnet 5 vs DeepSeek V4 Flash
Claude Sonnet 5 is the stronger model overall, scoring 54.6 to 53.6 on the Noometry Index. DeepSeek V4 Flash costs 15× less per token, which makes it the better buy when Claude Sonnet 5's lead doesn't matter for your workload.
Last verified . 37 shared benchmarks.
DeepSeekSummary
- They share 37 benchmarks with published results for both. Claude Sonnet 5 scores higher in 7 categories and DeepSeek V4 Flash in 1 category; 5 gaps are clear of the uncertainty.
- The widest gap is in coding, where Claude Sonnet 5 leads 55.5 to 47.9.
- The biggest single-benchmark swing is FrontierCode: 42.7% for Claude Sonnet 5 and 18.8% for DeepSeek V4 Flash.
- DeepSeek V4 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $2 / $10 for Claude Sonnet 5.
- DeepSeek V4 Flash has downloadable open weights; the other is API-only.
Side by side
| Claude Sonnet 5 | DeepSeek V4 Flash | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 54.6 | 53.6 |
| Released | 2026-06-29 | 2026-04-24 |
| Weights | Proprietary | Open |
| Context window | 1M | 1M |
| Max output | 128K | 393K |
| Input $ / M tokens | $2 | $0.15 |
| Output $ / M tokens | $10 | $0.60 |
| Results tracked | 51 | 41 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Claude Sonnet 5 leads
Claude Sonnet 5: 55.5 (#26), DeepSeek V4 Flash: 47.9 (#59)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| FrontierCode | 42.7% | 18.8% |
| LMArena WebDev | 1541 | 1582 |
| SciCode | 54.3% | 49.9% |
| WeirdML | 68.8% | 63% |
| LMArena Coding | 1483 | 1457 |
| ALE-Bench | 1,463 | 1,306 |
| DeepSWE | 53.8% | — |
| CursorBench | 34.1% | — |
| GSO | 37.3% | — |
Agentic & Tool Use Not comparable
Claude Sonnet 5: 42.8 (#18), DeepSeek V4 Flash: —
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| APEX-Agents | 54.5% | — |
| GBAEval | 65.3% | — |
| LMArena Search | 1194 | — |
| Vending-Bench 2 | 6,378 | — |
Reasoning DeepSeek V4 Flash leads
Claude Sonnet 5: 49.1 (#39), DeepSeek V4 Flash: 53.7 (#30)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| SimpleBench | 60.6% | 61.1% |
| NYT Connections (extended) | 75.1% | 89.6% |
| CritPt | 16.9% | 16.6% |
| Chess Puzzles | 35% | 33% |
| LMArena Hard Prompts | 1461 | 1444 |
| Mystery Game Puzzles | 35% | 34% |
| DTBench | 92.5% | 90.9% |
| LMCA | 50% | 41.7% |
| Epoch Capabilities Index | 156.21 | 154.49 |
| ARC-AGI-2 | — | 61.4% |
| Kagi LLM Benchmark | — | 52.2% |
| ARC-AGI-1 | — | 89% |
| Surface Evolver Bench | 60% | — |
| Bench to the Future 3 | 0.14 | — |
| ForecastBench | 61.1 | — |
Math Claude Sonnet 5 leads
Claude Sonnet 5: 66.2 (#27), DeepSeek V4 Flash: 60.3 (#37)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| FrontierMath (Tiers 1-3) | 65.6% | 57.5% |
| FrontierMath Tier 4 | 29.3% | 24.4% |
| OTIS Mock AIME 2024-2025 | 94.7% | 94.4% |
| ProofBench | 77% | 56% |
| LMArena Math | 1467 | 1427 |
| MathArena Final-Answer Competitions | — | 76.5% |
Knowledge Too close to call
Claude Sonnet 5: 55.6 (#47), DeepSeek V4 Flash: 55.4 (#48)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| GPQA Diamond | 90.5% | 91% |
| SimpleQA Verified | 33.7% | 33.6% |
| LMArena Expert | 1490 | 1441 |
Multimodal Not comparable
Claude Sonnet 5: 42.4 (#31), DeepSeek V4 Flash: —
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| LMArena Vision | 1274 | — |
| Blueprint-Bench 2 | 24.9% | — |
| LMArena Document | 1466 | — |
Multilingual Too close to call
Claude Sonnet 5: 53.8 (#55), DeepSeek V4 Flash: 53.0 (#72)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| LMArena Non-English | 1431 | 1420 |
| LMArena Chinese | 1477 | 1468 |
| LMArena French | 1460 | 1439 |
| LMArena German | 1440 | 1418 |
| LMArena Japanese | 1422 | 1406 |
| LMArena Korean | 1411 | 1384 |
| LMArena Russian | 1451 | 1428 |
| LMArena Spanish | 1437 | 1436 |
Instruction Following Claude Sonnet 5 leads
Claude Sonnet 5: 76.3 (#41), DeepSeek V4 Flash: 74.9 (#81)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| LMArena Instruction Following | 1452 | 1421 |
Long Context Too close to call
Claude Sonnet 5: 44.8 (#55), DeepSeek V4 Flash: 43.8 (#85)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| LMArena Longer Query | 1463 | 1434 |
Writing & Preference Claude Sonnet 5 leads
Claude Sonnet 5: 69.2 (#25), DeepSeek V4 Flash: 63.8 (#61)
| Benchmark | Claude Sonnet 5 | DeepSeek V4 Flash |
|---|---|---|
| LMArena Text | 1442 | 1432 |
| LMArena Creative Writing | 1416 | 1403 |
| EQ-Bench Creative Writing | 1794 | 1559 |
| LMArena Multi-Turn | 1454 | 1449 |
| EQ-Bench 4 | 1236 | — |
Frequently asked questions
Is Claude Sonnet 5 better than DeepSeek V4 Flash?
Claude Sonnet 5 is the stronger model overall, scoring 54.6 to 53.6 on the Noometry Index. DeepSeek V4 Flash costs 15× less per token, which makes it the better buy when Claude Sonnet 5's lead doesn't matter for your workload.
Which is cheaper, Claude Sonnet 5 or DeepSeek V4 Flash?
DeepSeek V4 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; Claude Sonnet 5 lists at $2 and $10.
Is Claude Sonnet 5 or DeepSeek V4 Flash better for coding?
Claude Sonnet 5 scores higher on coding benchmarks: 55.5 versus 47.9 in the Noometry coding category.
Which has the bigger context window?
Both accept 1M tokens.
How many benchmarks do Claude Sonnet 5 and DeepSeek V4 Flash share?
37 benchmarks have published results for both models. Claude Sonnet 5 has 51 scored results on Noometry and DeepSeek V4 Flash has 41.