Model comparison
DeepSeek-V3 vs GPT-5.4 nano
GPT-5.4 nano is the stronger model overall, scoring 41.9 to 39.5 on the Noometry Index.
Last verified . 29 shared benchmarks.
DeepSeekSummary
- They share 29 benchmarks with published results for both. DeepSeek-V3 scores higher in 2 categories and GPT-5.4 nano in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.4 nano leads 40.9 to 32.1.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 87.8% for GPT-5.4 nano.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.20 / $1.25 for GPT-5.4 nano.
- GPT-5.4 nano accepts more context: 400K tokens versus 164K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3 | GPT-5.4 nano | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.5 | 41.9 |
| Released | 2024-12-26 | 2026-03-17 |
| Weights | Open | Proprietary |
| Context window | 164K | 400K |
| Max output | 164K | 128K |
| Input $ / M tokens | $0.24 | $0.20 |
| Output $ / M tokens | $0.90 | $1.25 |
| Results tracked | 60 | 40 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 nano leads
DeepSeek-V3: 42.3 (#106), GPT-5.4 nano: 43.6 (#84)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| SciCode | 35.8% | 46.9% |
| WeirdML | 36.1% | 49.2% |
| LMArena Coding | 1368 | 1405 |
| Aider Polyglot | 55.1% | — |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| ALE-Bench | — | 1,005 |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GPT-5.4 nano: —
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| METR Time Horizons | 49.6% | — |
Reasoning GPT-5.4 nano leads
DeepSeek-V3: 20.5 (#236), GPT-5.4 nano: 23.7 (#173)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| Kagi LLM Benchmark | 52.3% | 39.7% |
| CritPt | 0% | 9.3% |
| LMArena Hard Prompts | 1365 | 1381 |
| DTBench | 64.8% | 80.3% |
| LMCA | 15.5% | 36.9% |
| Epoch Capabilities Index | 135.94 | 145.81 |
| ForecastBench | 59.1 | 57.3 |
| ARC-AGI-2 | — | 5.7% |
| SimpleBench | 27.2% | — |
| ARC-AGI-1 | — | 51.5% |
| Chess Puzzles | — | 30% |
| LiveBench Reasoning | 65.8% | — |
| Mystery Game Puzzles | — | 9% |
| LiveBench Data Analysis | 60.9% | — |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math GPT-5.4 nano leads
DeepSeek-V3: 32.1 (#219), GPT-5.4 nano: 40.9 (#88)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 87.8% |
| LMArena Math | 1373 | 1406 |
| FrontierMath (Feb 2025 set) | 1.7% | 25.9% |
| FrontierMath (Tiers 1-3) | — | 44.9% |
| FrontierMath Tier 4 | — | 12.2% |
| ProofBench | — | 5% |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath Tier 4 (v1) | — | 6.3% |
Knowledge GPT-5.4 nano leads
DeepSeek-V3: 37.5 (#155), GPT-5.4 nano: 41.9 (#103)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| GPQA Diamond | 67.6% | 78.5% |
| Vectara Hallucination Rate | 6.1% | 3.1% |
| LMArena Expert | 1351 | 1396 |
| SimpleQA Verified | — | 11.7% |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multimodal Not comparable
DeepSeek-V3: —, GPT-5.4 nano: 36.7 (#78)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| LMArena Vision | — | 1196 |
Multilingual Too close to call
DeepSeek-V3: 48.5 (#143), GPT-5.4 nano: 48.6 (#140)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| LMArena Non-English | 1358 | 1359 |
| LMArena Chinese | 1391 | 1392 |
| LMArena French | 1385 | 1396 |
| LMArena German | 1374 | 1367 |
| LMArena Japanese | 1333 | 1343 |
| LMArena Korean | 1319 | 1320 |
| LMArena Russian | 1373 | 1363 |
| LMArena Spanish | 1358 | 1371 |
Instruction Following Too close to call
DeepSeek-V3: 72.8 (#130), GPT-5.4 nano: 71.9 (#144)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| LMArena Instruction Following | 1345 | 1362 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context GPT-5.4 nano leads
DeepSeek-V3: 34.0 (#253), GPT-5.4 nano: 41.6 (#137)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| LMArena Longer Query | 1352 | 1366 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), GPT-5.4 nano: 55.7 (#142)
| Benchmark | DeepSeek-V3 | GPT-5.4 nano |
|---|---|---|
| LMArena Text | 1375 | 1372 |
| LMArena Creative Writing | 1364 | 1314 |
| LMArena Multi-Turn | 1389 | 1382 |
| Short-Story Creative Writing | 77% | — |
| EQ-Bench Creative Writing | 1472 | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than GPT-5.4 nano?
GPT-5.4 nano is the stronger model overall, scoring 41.9 to 39.5 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or GPT-5.4 nano?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; GPT-5.4 nano lists at $0.20 and $1.25.
Is DeepSeek-V3 or GPT-5.4 nano better for coding?
GPT-5.4 nano scores higher on coding benchmarks: 43.6 versus 42.3 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 nano does, with 400K tokens against 164K.
How many benchmarks do DeepSeek-V3 and GPT-5.4 nano share?
29 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-5.4 nano has 40.