Model comparison
DeepSeek V4 Pro vs GPT-5.6 Sol
GPT-5.6 Sol is the stronger model overall, scoring 65.0 to 54.3 on the Noometry Index. DeepSeek V4 Pro costs 8.1× less per token, which makes it the better buy when GPT-5.6 Sol's lead doesn't matter for your workload.
Last verified . 44 shared benchmarks.
DeepSeekSummary
- They share 44 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 0 categories and GPT-5.6 Sol in 9 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.6 Sol leads 85.6 to 64.8.
- The biggest single-benchmark swing is FrontierMath Tier 4: 26.8% for DeepSeek V4 Pro and 82.9% for GPT-5.6 Sol.
- DeepSeek V4 Pro is cheaper at $0.66 / $1.98 per million input/output tokens, against $4 / $20 for GPT-5.6 Sol.
- GPT-5.6 Sol accepts more context: 1.05M tokens versus 1M.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4 Pro | GPT-5.6 Sol | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 54.3 | 65.0 |
| Released | 2026-04-24 | 2026-07-09 |
| Weights | Open | Proprietary |
| Context window | 1M | 1.05M |
| Max output | 393K | 128K |
| Input $ / M tokens | $0.66 | $4 |
| Output $ / M tokens | $1.98 | $20 |
| Results tracked | 48 | 65 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.6 Sol leads
DeepSeek V4 Pro: 52.4 (#34), GPT-5.6 Sol: 65.1 (#7)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| FrontierCode | 28.6% | 47.5% |
| LMArena WebDev | 1582 | 1618 |
| SciCode | 51% | 57.1% |
| WeirdML | 66.2% | 89.4% |
| LMArena Coding | 1470 | 1498 |
| ALE-Bench | 1,403 | 2,177 |
| SWE-bench Verified | 77.6% | — |
| DeepSWE | — | 72.7% |
| CursorBench | — | 41.7% |
| FrontierSWE | — | 32.2% |
| GSO | — | 76.5% |
| MirrorCode | — | 20% |
Agentic & Tool Use GPT-5.6 Sol leads
DeepSeek V4 Pro: 32.8 (#58), GPT-5.6 Sol: 50.3 (#7)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| APEX-Agents | 47.3% | 51.4% |
| Vending-Bench 2 | 3,285 | 9,619 |
| OSWorld 2.0 | — | 27.3% |
| τ²-bench Banking | — | 46.9% |
| PostTrainBench | — | 36.2% |
| BALROG | — | 60% |
| GBAEval | — | 52.6% |
| GDP.pdf | — | 30.7% |
| LMArena Search | — | 1257 |
Reasoning GPT-5.6 Sol leads
DeepSeek V4 Pro: 56.5 (#24), GPT-5.6 Sol: 74.8 (#8)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| ARC-AGI-2 | 61.3% | 92.5% |
| Kagi LLM Benchmark | 53.5% | 67% |
| NYT Connections (extended) | 91.3% | 93.8% |
| ARC-AGI-1 | 90.5% | 97.5% |
| CritPt | 18% | 32.3% |
| Chess Puzzles | 47% | 64% |
| LMArena Hard Prompts | 1461 | 1484 |
| Mystery Game Puzzles | 43% | 58% |
| DTBench | 93.9% | 96% |
| LMCA | 45.5% | 59.2% |
| Surface Evolver Bench | 40% | 93.1% |
| Epoch Capabilities Index | 155.31 | 161.66 |
| SimpleBench | — | 71.7% |
| EnigmaEval | — | 37.1% |
| EBR-Bench | — | 44.8% |
| Bench to the Future 3 | — | 0.14 |
| ForecastBench | 56.1 | — |
Math GPT-5.6 Sol leads
DeepSeek V4 Pro: 64.8 (#30), GPT-5.6 Sol: 85.6 (#9)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| FrontierMath (Tiers 1-3) | 64.6% | 89.1% |
| FrontierMath Tier 4 | 26.8% | 82.9% |
| OTIS Mock AIME 2024-2025 | 98.6% | 100% |
| ProofBench | 50% | 83% |
| LMArena Math | 1455 | 1474 |
| MathArena Final-Answer Competitions | 76.6% | — |
| FrontierMath Erdős | — | 0% |
Knowledge GPT-5.6 Sol leads
DeepSeek V4 Pro: 59.5 (#31), GPT-5.6 Sol: 64.3 (#18)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| GPQA Diamond | 91.7% | 93.5% |
| SimpleQA Verified | 52.9% | 69.7% |
| Vectara Hallucination Rate | 8.6% | 12.4% |
| LMArena Expert | 1464 | 1516 |
Multimodal Not comparable
DeepSeek V4 Pro: —, GPT-5.6 Sol: 48.6 (#9)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| LMArena Vision | — | 1281 |
| Blueprint-Bench 2 | — | 33.6% |
| Furniture Assembly | — | 56.7% |
| LMArena Document | — | 1483 |
Multilingual Too close to call
DeepSeek V4 Pro: 54.4 (#45), GPT-5.6 Sol: 55.3 (#32)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| LMArena Non-English | 1439 | 1452 |
| LMArena Chinese | 1486 | 1527 |
| LMArena French | 1472 | 1477 |
| LMArena German | 1458 | 1476 |
| LMArena Japanese | 1445 | 1471 |
| LMArena Korean | 1447 | 1442 |
| LMArena Russian | 1453 | 1468 |
| LMArena Spanish | 1458 | 1441 |
Instruction Following GPT-5.6 Sol leads
DeepSeek V4 Pro: 76.1 (#47), GPT-5.6 Sol: 77.7 (#16)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| LMArena Instruction Following | 1448 | 1482 |
Long Context Too close to call
DeepSeek V4 Pro: 45.0 (#51), GPT-5.6 Sol: 45.4 (#42)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| LMArena Longer Query | 1458 | 1480 |
| CL-bench Life | 13.5% | — |
Writing & Preference GPT-5.6 Sol leads
DeepSeek V4 Pro: 65.5 (#46), GPT-5.6 Sol: 73.3 (#12)
| Benchmark | DeepSeek V4 Pro | GPT-5.6 Sol |
|---|---|---|
| LMArena Text | 1451 | 1457 |
| LMArena Creative Writing | 1446 | 1448 |
| EQ-Bench Creative Writing | 1553 | 1972 |
| EQ-Bench 4 | 1166 | 1250 |
| LMArena Multi-Turn | 1467 | 1460 |
Frequently asked questions
Is DeepSeek V4 Pro better than GPT-5.6 Sol?
GPT-5.6 Sol is the stronger model overall, scoring 65.0 to 54.3 on the Noometry Index. DeepSeek V4 Pro costs 8.1× less per token, which makes it the better buy when GPT-5.6 Sol's lead doesn't matter for your workload.
Which is cheaper, DeepSeek V4 Pro or GPT-5.6 Sol?
DeepSeek V4 Pro is cheaper. It lists at $0.66 per million input tokens and $1.98 per million output tokens; GPT-5.6 Sol lists at $4 and $20.
Is DeepSeek V4 Pro or GPT-5.6 Sol better for coding?
GPT-5.6 Sol scores higher on coding benchmarks: 65.1 versus 52.4 in the Noometry coding category.
Which has the bigger context window?
GPT-5.6 Sol does, with 1.05M tokens against 1M.
How many benchmarks do DeepSeek V4 Pro and GPT-5.6 Sol share?
44 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and GPT-5.6 Sol has 65.