Model comparison
DeepSeek V4.1 Flash vs GPT-5.4
GPT-5.4 is the stronger model overall, scoring 59.4 to 52.8 on the Noometry Index. DeepSeek V4.1 Flash costs 21× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 35 shared benchmarks.
DeepSeekSummary
- They share 35 benchmarks with published results for both. DeepSeek V4.1 Flash scores higher in 2 categories and GPT-5.4 in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in agentic & tool use, where GPT-5.4 leads 46.5 to 31.2.
- The biggest single-benchmark swing is FrontierMath Tier 4: 26.8% for DeepSeek V4.1 Flash and 49% for GPT-5.4.
- DeepSeek V4.1 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 1M.
- DeepSeek V4.1 Flash has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4.1 Flash | GPT-5.4 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 52.8 | 59.4 |
| Released | 2026-09-09 | 2026-03-05 |
| Weights | Open | Proprietary |
| Context window | 1M | 1.05M |
| Max output | 393K | 128K |
| Input $ / M tokens | $0.15 | $2.50 |
| Output $ / M tokens | $0.60 | $15 |
| Results tracked | 37 | 68 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek V4.1 Flash: 52.9 (#32), GPT-5.4: 52.6 (#33)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| LMArena WebDev | 1619 | 1465 |
| SciCode | 51.9% | 56.6% |
| LMArena Coding | 1506 | 1497 |
| ALE-Bench | 1,092 | 1,607 |
| SWE-bench Verified | — | 76.9% |
| DeepSWE | — | 51.8% |
| GSO | — | 31.4% |
| WeirdML | — | 77.7% |
| MirrorCode | — | 15.6% |
| AlgoTune | — | 1.85 |
Agentic & Tool Use GPT-5.4 leads
DeepSeek V4.1 Flash: 31.2 (#69), GPT-5.4: 46.5 (#13)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| APEX-Agents | 39.5% | 52.4% |
| Terminal-Bench | — | 81.8% |
| τ²-bench Banking | — | 39.4% |
| DeepResearch Bench | — | 35.1% |
| PostTrainBench | — | 19% |
| GBAEval | — | 45.1% |
| GDP.pdf | 19.8% | — |
| LMArena Search | — | 1197 |
| METR Time Horizons | — | 74.3% |
| Vending-Bench 2 | — | 6,144 |
Reasoning GPT-5.4 leads
DeepSeek V4.1 Flash: 50.2 (#36), GPT-5.4: 61.8 (#19)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| NYT Connections (extended) | 89.6% | 91.3% |
| CritPt | 14.3% | 23.4% |
| LMArena Hard Prompts | 1483 | 1485 |
| Mystery Game Puzzles | 43% | 37% |
| DTBench | 89.9% | 94.4% |
| LMCA | 47% | 52% |
| Epoch Capabilities Index | 154.9 | 156.81 |
| ARC-AGI-2 | — | 74% |
| Kagi LLM Benchmark | — | 63.8% |
| ARC-AGI-1 | — | 93.7% |
| Chess Puzzles | — | 44% |
| EnigmaEval | — | 16% |
| Thematic Generalization | — | 80% |
| EBR-Bench | — | 25.4% |
| Surface Evolver Bench | 46.3% | — |
| ForecastBench | — | 59.5 |
Math GPT-5.4 leads
DeepSeek V4.1 Flash: 66.7 (#25), GPT-5.4: 73.5 (#19)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 67.4% | 78.6% |
| FrontierMath Tier 4 | 26.8% | 49% |
| OTIS Mock AIME 2024-2025 | 98.3% | 97.8% |
| ProofBench | 54% | 56% |
| LMArena Math | 1477 | 1488 |
| MathArena Final-Answer Competitions | — | 83.1% |
| FrontierMath (Feb 2025 set) | — | 47.6% |
| FrontierMath Tier 4 (v1) | — | 27.1% |
Knowledge GPT-5.4 leads
DeepSeek V4.1 Flash: 57.9 (#38), GPT-5.4: 65.3 (#14)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| GPQA Diamond | 89.8% | 93.3% |
| LMArena Expert | 1506 | 1507 |
| Humanity's Last Exam | — | 36.2% |
| SimpleQA Verified | — | 45.1% |
| Vectara Hallucination Rate | — | 7% |
Multimodal GPT-5.4 leads
DeepSeek V4.1 Flash: 39.1 (#61), GPT-5.4: 43.7 (#20)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| LMArena Vision | 1277 | 1303 |
| Furniture Assembly | 34.2% | 37.5% |
| Blueprint-Bench 2 | — | 27.1% |
| LMArena Document | — | 1471 |
Multilingual GPT-5.4 leads
DeepSeek V4.1 Flash: 55.0 (#35), GPT-5.4: 56.2 (#23)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1448 | 1465 |
| LMArena Chinese | 1497 | 1519 |
| LMArena French | 1452 | 1493 |
| LMArena German | 1484 | 1472 |
| LMArena Japanese | 1412 | 1485 |
| LMArena Korean | 1452 | 1448 |
| LMArena Russian | 1471 | 1480 |
| LMArena Spanish | 1459 | 1454 |
Instruction Following Too close to call
DeepSeek V4.1 Flash: 77.3 (#26), GPT-5.4: 77.1 (#27)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1474 | 1469 |
Long Context GPT-5.4 leads
DeepSeek V4.1 Flash: 45.2 (#47), GPT-5.4: 50.3 (#8)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| LMArena Longer Query | 1475 | 1473 |
| CL-bench | — | 27.9% |
| CL-bench Life | — | 21.7% |
Writing & Preference GPT-5.4 leads
DeepSeek V4.1 Flash: 65.4 (#48), GPT-5.4: 71.9 (#17)
| Benchmark | DeepSeek V4.1 Flash | GPT-5.4 |
|---|---|---|
| LMArena Text | 1462 | 1469 |
| LMArena Creative Writing | 1435 | 1439 |
| EQ-Bench Creative Writing | 1540 | 1840 |
| LMArena Multi-Turn | 1457 | 1482 |
| EQ-Bench 4 | — | 1272 |
Frequently asked questions
Is DeepSeek V4.1 Flash better than GPT-5.4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 52.8 on the Noometry Index. DeepSeek V4.1 Flash costs 21× 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 V4.1 Flash or GPT-5.4?
DeepSeek V4.1 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is DeepSeek V4.1 Flash or GPT-5.4 better for coding?
They score almost the same on coding (52.9 vs 52.6); test both on your own repository before choosing.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 1M.
How many benchmarks do DeepSeek V4.1 Flash and GPT-5.4 share?
35 benchmarks have published results for both models. DeepSeek V4.1 Flash has 37 scored results on Noometry and GPT-5.4 has 68.