Model comparison
DeepSeek-V3.1 vs GPT-5.4
GPT-5.4 is the stronger model overall, scoring 59.4 to 42.8 on the Noometry Index. DeepSeek-V3.1 costs 13× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 25 shared benchmarks.
DeepSeekSummary
- They share 25 benchmarks with published results for both. DeepSeek-V3.1 scores higher in 0 categories and GPT-5.4 in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.4 leads 73.5 to 38.9.
- The biggest single-benchmark swing is WeirdML: 38.4% for DeepSeek-V3.1 and 77.7% for GPT-5.4.
- DeepSeek-V3.1 is cheaper at $0.25 / $0.95 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 164K.
- DeepSeek-V3.1 has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3.1 | GPT-5.4 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 42.8 | 59.4 |
| Released | 2025-08-21 | 2026-03-05 |
| Weights | Open | Proprietary |
| Context window | 164K | 1.05M |
| Max output | 8K | 128K |
| Input $ / M tokens | $0.25 | $2.50 |
| Output $ / M tokens | $0.95 | $15 |
| Results tracked | 27 | 68 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 leads
DeepSeek-V3.1: 40.3 (#144), GPT-5.4: 52.6 (#33)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| WeirdML | 38.4% | 77.7% |
| LMArena Coding | 1417 | 1497 |
| SWE-bench Verified | — | 76.9% |
| DeepSWE | — | 51.8% |
| LMArena WebDev | — | 1465 |
| SciCode | — | 56.6% |
| GSO | — | 31.4% |
| MirrorCode | — | 15.6% |
| ALE-Bench | — | 1,607 |
| AlgoTune | — | 1.85 |
Agentic & Tool Use Not comparable
DeepSeek-V3.1: —, GPT-5.4: 46.5 (#13)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| Terminal-Bench | — | 81.8% |
| APEX-Agents | — | 52.4% |
| τ²-bench Banking | — | 39.4% |
| DeepResearch Bench | — | 35.1% |
| PostTrainBench | — | 19% |
| GBAEval | — | 45.1% |
| LMArena Search | — | 1197 |
| METR Time Horizons | — | 74.3% |
| Vending-Bench 2 | — | 6,144 |
Reasoning GPT-5.4 leads
DeepSeek-V3.1: 27.9 (#110), GPT-5.4: 61.8 (#19)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| Kagi LLM Benchmark | 53.2% | 63.8% |
| LMArena Hard Prompts | 1417 | 1485 |
| DTBench | 82.7% | 94.4% |
| LMCA | 24.3% | 52% |
| Epoch Capabilities Index | 139.92 | 156.81 |
| ForecastBench | 58 | 59.5 |
| ARC-AGI-2 | — | 74% |
| SimpleBench | 40% | — |
| NYT Connections (extended) | — | 91.3% |
| ARC-AGI-1 | — | 93.7% |
| CritPt | — | 23.4% |
| Chess Puzzles | — | 44% |
| EnigmaEval | — | 16% |
| Thematic Generalization | — | 80% |
| EBR-Bench | — | 25.4% |
| Mystery Game Puzzles | — | 37% |
Math GPT-5.4 leads
DeepSeek-V3.1: 38.9 (#122), GPT-5.4: 73.5 (#19)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| LMArena Math | 1420 | 1488 |
| FrontierMath (Tiers 1-3) | — | 78.6% |
| FrontierMath Tier 4 | — | 49% |
| MathArena Final-Answer Competitions | — | 83.1% |
| OTIS Mock AIME 2024-2025 | — | 97.8% |
| ProofBench | — | 56% |
| FrontierMath (Feb 2025 set) | — | 47.6% |
| FrontierMath Tier 4 (v1) | — | 27.1% |
Knowledge GPT-5.4 leads
DeepSeek-V3.1: 43.7 (#90), GPT-5.4: 65.3 (#14)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| Vectara Hallucination Rate | 5.5% | 7% |
| LMArena Expert | 1405 | 1507 |
| GPQA Diamond | — | 93.3% |
| Humanity's Last Exam | — | 36.2% |
| SimpleQA Verified | — | 45.1% |
Multimodal Not comparable
DeepSeek-V3.1: —, GPT-5.4: 43.7 (#20)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| LMArena Vision | — | 1303 |
| Blueprint-Bench 2 | — | 27.1% |
| Furniture Assembly | — | 37.5% |
| LMArena Document | — | 1471 |
Multilingual GPT-5.4 leads
DeepSeek-V3.1: 51.6 (#106), GPT-5.4: 56.2 (#23)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1400 | 1465 |
| LMArena Chinese | 1469 | 1519 |
| LMArena French | 1447 | 1493 |
| LMArena German | 1411 | 1472 |
| LMArena Japanese | 1378 | 1485 |
| LMArena Korean | 1337 | 1448 |
| LMArena Russian | 1405 | 1480 |
| LMArena Spanish | 1431 | 1454 |
Instruction Following GPT-5.4 leads
DeepSeek-V3.1: 73.9 (#110), GPT-5.4: 77.1 (#27)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1400 | 1469 |
Long Context GPT-5.4 leads
DeepSeek-V3.1: 36.3 (#232), GPT-5.4: 50.3 (#8)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| LMArena Longer Query | 1422 | 1473 |
| Fiction.LiveBench | 52.8% | — |
| CL-bench | — | 27.9% |
| CL-bench Life | — | 21.7% |
Writing & Preference GPT-5.4 leads
DeepSeek-V3.1: 60.3 (#98), GPT-5.4: 71.9 (#17)
| Benchmark | DeepSeek-V3.1 | GPT-5.4 |
|---|---|---|
| LMArena Text | 1420 | 1469 |
| LMArena Creative Writing | 1401 | 1439 |
| EQ-Bench Creative Writing | 1436 | 1840 |
| LMArena Multi-Turn | 1408 | 1482 |
| EQ-Bench 4 | — | 1272 |
Frequently asked questions
Is DeepSeek-V3.1 better than GPT-5.4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 42.8 on the Noometry Index. DeepSeek-V3.1 costs 13× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.1 or GPT-5.4?
DeepSeek-V3.1 is cheaper. It lists at $0.25 per million input tokens and $0.95 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is DeepSeek-V3.1 or GPT-5.4 better for coding?
GPT-5.4 scores higher on coding benchmarks: 52.6 versus 40.3 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 164K.
How many benchmarks do DeepSeek-V3.1 and GPT-5.4 share?
25 benchmarks have published results for both models. DeepSeek-V3.1 has 27 scored results on Noometry and GPT-5.4 has 68.