Model comparison
DeepSeek-V3 vs GPT-3.5-turbo
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 23.2 on the Noometry Index.
Last verified . 36 shared benchmarks.
DeepSeekSummary
- They share 36 benchmarks with published results for both. DeepSeek-V3 scores higher in 8 categories and GPT-3.5-turbo in 0 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3 leads 57.4 to 25.3.
- The biggest single-benchmark swing is MATH Level 5: 75.5% for DeepSeek-V3 and 15.9% for GPT-3.5-turbo.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.50 / $1.50 for GPT-3.5-turbo.
- DeepSeek-V3 accepts more context: 164K tokens versus 16K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3 | GPT-3.5-turbo | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.5 | 23.2 |
| Released | 2024-12-26 | 2023-03-01 |
| Weights | Open | Proprietary |
| Context window | 164K | 16K |
| Max output | 164K | 4K |
| Input $ / M tokens | $0.24 | $0.50 |
| Output $ / M tokens | $0.90 | $1.50 |
| Results tracked | 60 | 44 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V3: 42.3 (#106), GPT-3.5-turbo: 23.9 (#331)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| WeirdML | 36.1% | 3.5% |
| BigCodeBench Instruct | 50% | 39.1% |
| LMArena Coding | 1368 | 1136 |
| BigCodeBench Complete | 62.2% | 50.6% |
| HumanEval+ | 86.6% | 70.7% |
| MBPP+ | 73% | 69.7% |
| Aider Polyglot | 55.1% | — |
| SciCode | 35.8% | — |
| LiveBench Coding | 70.9% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GPT-3.5-turbo: —
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| METR Time Horizons | 49.6% | 21.5% |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.5 (#236), GPT-3.5-turbo: 13.8 (#332)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| LMArena Hard Prompts | 1365 | 1108 |
| DTBench | 64.8% | 48.5% |
| LMCA | 15.5% | 9.7% |
| BIG-Bench Hard | 87.5% | 61.6% |
| Epoch Capabilities Index | 135.94 | 118.55 |
| ForecastBench | 59.1 | 50.4 |
| WinoGrande | 85.2% | 81.6% |
| SimpleBench | 27.2% | — |
| Kagi LLM Benchmark | 52.3% | — |
| CritPt | 0% | — |
| Chess Puzzles | — | 0% |
| LiveBench Reasoning | 65.8% | — |
| Mystery Game Puzzles | — | 3% |
| LiveBench Data Analysis | 60.9% | — |
| Adversarial NLI | — | 58.1% |
| CommonsenseQA 2.0 | — | 57% |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
Math DeepSeek-V3 leads
DeepSeek-V3: 32.1 (#219), GPT-3.5-turbo: 6.3 (#327)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 2.2% |
| LMArena Math | 1373 | 1142 |
| MATH Level 5 | 75.5% | 15.9% |
| FrontierMath (Tiers 1-3) | — | 0% |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
| GSM8K | — | 57.8% |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), GPT-3.5-turbo: 10.0 (#303)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| GPQA Diamond | 67.6% | 28% |
| LMArena Expert | 1351 | 1070 |
| ARC (AI2) Challenge | 95.3% | 87.4% |
| MMLU | 87.2% | 71.4% |
| TriviaQA | 82.9% | 85.8% |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| Vectara Hallucination Rate | 6.1% | — |
| GPQA (HELM) | 53.8% | — |
| BoolQ | — | 87% |
| OpenBookQA | — | 86% |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.5 (#143), GPT-3.5-turbo: 31.5 (#258)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| LMArena Non-English | 1358 | 1108 |
| LMArena Chinese | 1391 | 1075 |
| LMArena French | 1385 | 1118 |
| LMArena German | 1374 | 1090 |
| LMArena Japanese | 1333 | 1043 |
| LMArena Korean | 1319 | 1019 |
| LMArena Russian | 1373 | 1123 |
| LMArena Spanish | 1358 | 1121 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), GPT-3.5-turbo: 57.9 (#262)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| LMArena Instruction Following | 1345 | 1119 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context Too close to call
DeepSeek-V3: 34.0 (#253), GPT-3.5-turbo: 34.0 (#254)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| LMArena Longer Query | 1352 | 1121 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), GPT-3.5-turbo: 25.3 (#305)
| Benchmark | DeepSeek-V3 | GPT-3.5-turbo |
|---|---|---|
| LMArena Text | 1375 | 1125 |
| LMArena Creative Writing | 1364 | 1092 |
| EQ-Bench Creative Writing | 1472 | 451 |
| LMArena Multi-Turn | 1389 | 1117 |
| Short-Story Creative Writing | 77% | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than GPT-3.5-turbo?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 23.2 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or GPT-3.5-turbo?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; GPT-3.5-turbo lists at $0.50 and $1.50.
Is DeepSeek-V3 or GPT-3.5-turbo better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 23.9 in the Noometry coding category.
Which has the bigger context window?
DeepSeek-V3 does, with 164K tokens against 16K.
How many benchmarks do DeepSeek-V3 and GPT-3.5-turbo share?
36 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-3.5-turbo has 44.