DeepSeek, open weights
DeepSeek-V3
DeepSeek-V3 by DeepSeek ranks 166th of 354 ranked models on the Noometry Index as of October 2026, with a score of 39.5. Its strongest category is coding, where it ranks 106th. API pricing starts at $0.24 per million input tokens and $0.90 per million output tokens, with a 164K-token context window.
Last verified
Specifications
- Noometry rank
- #166 of 354
- Index score
- 39.5
- Evidence
- Confirmed 60 results
- Provider
DeepSeek- Released
- December 26, 2024
- Weights
- Open weights
- Reasoning
- No
- Context window
- 164K
- Max output
- 164K
- Input price
- $0.24 / M
- Output price
- $0.90 / M
- Blended price
- $0.41 / M
- Output speed
- 73 tokens/s Kagi
- Value
- #62 of 219
- Knowledge cutoff
- July 2024
- Input
- text
- Hugging Face
- deepseek-ai/DeepSeek-V3
Category scores
Each category score combines every public result we have in that category.
- Coding 42.3
- Reasoning 20.5
- Math 32.1
- Knowledge 37.5
- Multilingual 48.5
- Instruction Following 72.8
- Long Context 34.0
- Writing & Preference 57.4
| Category | Score | Rank | Results |
|---|---|---|---|
| Coding | 42.3 | #106 | 7 |
| Reasoning | 20.5 | #236 | 8 |
| Math | 32.1 | #219 | 5 |
| Knowledge | 37.5 | #155 | 6 |
| Multilingual | 48.5 | #143 | 1 |
| Instruction Following | 72.8 | #130 | 3 |
| Long Context | 34.0 | #253 | 2 |
| Writing & Preference | 57.4 | #130 | 7 |
Strengths and weaknesses
Categories where DeepSeek-V3 places highest and lowest among the models ranked in each, with its score against that category's median.
Strongest categories
| Category | Score | vs median | Rank |
|---|---|---|---|
| Coding | 42.3 | +3.6 | #106 of 340, top 32% |
| Writing & Preference | 57.4 | +3.6 | #130 of 312, top 42% |
| Instruction Following | 72.8 | +1.5 | #130 of 305, top 43% |
Weakest categories
| Category | Score | vs median | Rank |
|---|---|---|---|
| Long Context | 34.0 | −6.9 | #253 of 296, top 86% |
| Reasoning | 20.5 | −3.1 | #236 of 350, top 68% |
| Math | 32.1 | −4.5 | #219 of 327, top 67% |
Closest competitors
The models ranked just above and below DeepSeek-V3. When scores are this close, price and speed are often the better way to choose.
| Model | Rank | Score | Blended $/M | Speed | |
|---|---|---|---|---|---|
| DeepSeek-V3.2-Speciale | #162 | 39.7 | $0.85 | — | Compare |
| Hunyuan Turbo 0110 | #163 | 39.6 | — | — | Compare |
| Claude 3.7 Sonnet | #164 | 39.5 | — | — | Compare |
| Claude Haiku 4.5 | #165 | 39.5 | $2 | — | Compare |
| Grok 4 Fast | #167 | 39.4 | — | 577 | Compare |
| Olmo 3.1 32b Instruct | #168 | 39.4 | — | — | Compare |
| Granite 4.2 3b | #169 | 39.4 | — | — | Compare |
| Gemini 2.5 Flash | #170 | 39.3 | $0.85 | 152 | Compare |
Sponsored placements are available on pages like this one. Advertise on Noometry
Benchmark results
Every published result we track, with its source. Bold rows are the ones used for ranking; where several exist we prefer independent runs over self-reported numbers.
Coding
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| Aider Polyglot | 55.1% | #18 of 44, top 41% | Epoch AI | ||
| Aider Polyglot | 48.4% | Epoch AI | |||
| SciCode | 35.4% | Epoch AI | |||
| SciCode | 35.8% | #95 of 121, top 79% | Epoch AI | ||
| WeirdML | 36.1% | #90 of 119, top 76% | Epoch AI | ||
| BigCodeBench Instruct | 50% | #2 of 64, top 4% | BigCodeBench | 2024-12-26 | |
| LiveBench Coding | 70.9% | #7 of 39, top 18% | Epoch AI | ||
| LiveBench Coding | 61.8% | Epoch AI | |||
| LMArena Coding | 1325 | LMArena | 2026-10-08 | ||
| LMArena Coding | 1368 | #159 of 294, top 55% | LMArena | 2026-10-08 | |
| BigCodeBench Complete | 62.2% | Best of 66 | BigCodeBench | 2024-12-26 | |
| HumanEval+ | 86.6% | #5 of 45, top 12% | nov 2024 | EvalPlus | |
| MBPP+ | 73% | #10 of 38, top 27% | nov 2024 | EvalPlus |
Agentic & Tool Use
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| METR Time Horizons | 47.4% | Epoch AI | |||
| METR Time Horizons | 49.6% | #24 of 32, top 75% | Epoch AI |
Reasoning
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| SimpleBench | 18.9% | Epoch AI | |||
| SimpleBench | 27.2% | #62 of 77, top 81% | Epoch AI | ||
| Kagi LLM Benchmark | 52.3% | #58 of 99, top 59% | Kagi LLM Benchmark | ||
| CritPt | 0% | #103 of 134, top 77% | Epoch AI | ||
| CritPt | 0% | #103 of 134, top 77% | Epoch AI | ||
| LiveBench Reasoning | 65.8% | #12 of 39, top 31% | Epoch AI | ||
| LiveBench Reasoning | 56.8% | Epoch AI | |||
| LMArena Hard Prompts | 1312 | LMArena | 2026-10-08 | ||
| LMArena Hard Prompts | 1365 | #150 of 297, top 51% | LMArena | 2026-10-08 | |
| DTBench | 64.8% | #102 of 151, top 68% | Epoch AI | ||
| LiveBench Data Analysis | 60.4% | Epoch AI | |||
| LiveBench Data Analysis | 60.9% | #13 of 39, top 34% | Epoch AI | ||
| LMCA | 15.5% | #104 of 125, top 84% | Epoch AI | ||
| BIG-Bench Hard | 87.5% | #2 of 27, top 8% | Epoch AI | ||
| Epoch Capabilities Index | 135.94 | #122 of 213, top 58% | Epoch AI | 2025-03-24 | |
| Epoch Capabilities Index | 132.34 | Epoch AI | 2024-12-26 | ||
| ForecastBench | 59.1 | #41 of 72, top 57% | Epoch AI | ||
| HellaSwag | 88.9% | #5 of 29, top 18% | Epoch AI | ||
| LiveBench | 60.5% | Epoch AI | |||
| LiveBench | 66.9% | #10 of 39, top 26% | Epoch AI | ||
| PIQA | 84.7% | #6 of 27, top 23% | Epoch AI | ||
| WinoGrande | 85.2% | #7 of 43, top 17% | Epoch AI |
Math
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | #117 of 173, top 68% | Epoch AI | 2025-04-01 | |
| OTIS Mock AIME 2024-2025 | 15.8% | Epoch AI | 2025-02-25 | ||
| Omni-MATH | 40.3% | #26 of 57, top 46% | HELM Capabilities | ||
| LiveBench Math | 73.5% | #9 of 39, top 24% | Epoch AI | ||
| LiveBench Math | 60.5% | Epoch AI | |||
| LMArena Math | 1373 | #148 of 285, top 52% | LMArena | 2026-10-08 | |
| LMArena Math | 1311 | LMArena | 2026-10-08 | ||
| MATH Level 5 | 64.9% | Epoch AI | 2025-01-27 | ||
| MATH Level 5 | 75.5% | #27 of 79, top 35% | Epoch AI | 2025-04-01 | |
| FrontierMath (Feb 2025 set) | 1.7% | #55 of 68, top 81% | Epoch AI | 2025-03-07 |
Knowledge
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| GPQA Diamond | 67.6% | #101 of 186, top 55% | Epoch AI | 2025-04-01 | |
| GPQA Diamond | 56.5% | Epoch AI | 2025-01-27 | ||
| MMLU-Pro | 72.3% | #32 of 58, top 56% | HELM Capabilities | ||
| Confabulations (lower is better) | 26.1% | #43 of 51, top 85% | Lech Mazur benchmarks | ||
| Vectara Hallucination Rate (lower is better) | 6.1% | #23 of 96, top 24% | Vectara Hallucination Leaderboard | ||
| GPQA (HELM) | 53.8% | #28 of 57, top 50% | HELM Capabilities | ||
| LMArena Expert | 1306 | LMArena | 2026-10-08 | ||
| LMArena Expert | 1351 | #156 of 273, top 58% | LMArena | 2026-10-08 | |
| ARC (AI2) Challenge | 95.3% | Best of 39 | Epoch AI | ||
| MMLU | 87.2% | #3 of 81, top 4% | Epoch AI | ||
| TriviaQA | 82.9% | #7 of 25, top 29% | Epoch AI |
Multilingual
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Non-English | 1358 | #143 of 297, top 49% | LMArena | 2026-10-08 | |
| LMArena Non-English | 1316 | LMArena | 2026-10-08 | ||
| LMArena Chinese | 1338 | LMArena | 2026-10-08 | ||
| LMArena Chinese | 1391 | #143 of 285, top 51% | LMArena | 2026-10-08 | |
| LMArena French | 1343 | LMArena | 2026-10-08 | ||
| LMArena French | 1385 | #129 of 223, top 58% | LMArena | 2026-10-08 | |
| LMArena German | 1324 | LMArena | 2026-10-08 | ||
| LMArena German | 1374 | #114 of 231, top 50% | LMArena | 2026-10-08 | |
| LMArena Japanese | 1266 | LMArena | 2026-10-08 | ||
| LMArena Japanese | 1333 | #108 of 211, top 52% | LMArena | 2026-10-08 | |
| LMArena Korean | 1319 | #120 of 213, top 57% | LMArena | 2026-10-08 | |
| LMArena Korean | 1248 | LMArena | 2026-10-08 | ||
| LMArena Russian | 1373 | #133 of 283, top 47% | LMArena | 2026-10-08 | |
| LMArena Russian | 1324 | LMArena | 2026-10-08 | ||
| LMArena Spanish | 1358 | #137 of 226, top 61% | LMArena | 2026-10-08 | |
| LMArena Spanish | 1352 | LMArena | 2026-10-08 |
Instruction Following
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LiveBench Instruction Following | 75.3% | Epoch AI | |||
| LiveBench Instruction Following | 81.5% | #8 of 39, top 21% | Epoch AI | ||
| IFEval | 83.2% | #30 of 57, top 53% | HELM Capabilities | ||
| LMArena Instruction Following | 1315 | LMArena | 2026-10-08 | ||
| LMArena Instruction Following | 1345 | #149 of 298, top 50% | LMArena | 2026-10-08 |
Long Context
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| Fiction.LiveBench | 50% | #34 of 47, top 73% | Epoch AI | ||
| LMArena Longer Query | 1352 | #152 of 291, top 53% | LMArena | 2026-10-08 | |
| LMArena Longer Query | 1343 | LMArena | 2026-10-08 |
Writing & Preference
| Benchmark | Score | Position | Setting | Source | Date |
|---|---|---|---|---|---|
| LMArena Text | 1333 | LMArena | 2026-10-08 | ||
| LMArena Text | 1375 | #141 of 297, top 48% | LMArena | 2026-10-08 | |
| LMArena Creative Writing | 1329 | LMArena | 2026-10-08 | ||
| LMArena Creative Writing | 1364 | #116 of 295, top 40% | LMArena | 2026-10-08 | |
| Short-Story Creative Writing | 77% | #19 of 39, top 49% | Epoch AI | ||
| EQ-Bench Creative Writing | 1472 | #62 of 115, top 54% | EQ-Bench | ||
| WildBench | 83% | #18 of 57, top 32% | HELM Capabilities | ||
| LMArena Multi-Turn | 1389 | #130 of 295, top 45% | LMArena | 2026-10-08 | |
| LMArena Multi-Turn | 1349 | LMArena | 2026-10-08 | ||
| LiveBench Language | 49.1% | #12 of 39, top 31% | Epoch AI | ||
| LiveBench Language | 47.5% | Epoch AI |
API pricing by provider
| Route | Input $/M | Output $/M | Cached input $/M | Checked |
|---|---|---|---|---|
| deepinfra | $0.24 | $0.90 | $0.14 | 2026-10-10 |
| openrouter | $0.32 | $0.89 | — | 2026-10-10 |
| together | $1.25 | $1.25 | — | 2026-10-10 |
Compare DeepSeek-V3
- DeepSeek-V3 vs Claude Haiku 4.5
- DeepSeek-V3 vs Grok 4 Fast
- DeepSeek-V3 vs Claude 3.7 Sonnet
- DeepSeek-V3 vs Olmo 3.1 32b Instruct
- DeepSeek-V3 vs Hunyuan Turbo 0110
- DeepSeek-V3 vs Granite 4.2 3b
- DeepSeek-V3 vs GPT-6 Astra
- DeepSeek-V3 vs Claude Fable 5.1
- DeepSeek-V3 vs Gemini 3.8 Flash
- DeepSeek-V3 vs Kimi K3
- DeepSeek-V3 vs Grok 4.6
- DeepSeek-V3 vs Qwen3.8 Max
- DeepSeek-V3 vs GLM-5.3
- DeepSeek-V3 vs Muse Spark 1.3
Other DeepSeek models
Frequently asked questions
How good is DeepSeek-V3?
DeepSeek-V3 by DeepSeek ranks 166th of 354 ranked models on the Noometry Index as of October 2026, with a score of 39.5. Its strongest category is coding, where it ranks 106th. API pricing starts at $0.24 per million input tokens and $0.90 per million output tokens, with a 164K-token context window.
How much does DeepSeek-V3 cost?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens on deepinfra, with cached input at $0.14.
What is DeepSeek-V3's context window?
DeepSeek-V3 accepts up to 164K tokens of input and can write up to 164K tokens in one response.
Is DeepSeek-V3 open source?
Yes. DeepSeek-V3's weights are downloadable from Hugging Face (deepseek-ai/DeepSeek-V3); check the license for commercial terms.
How fast is DeepSeek-V3?
DeepSeek-V3 generated about 73 output tokens per second in the Kagi LLM Benchmark's timed runs. Speed varies by provider, load and reasoning effort.
What are DeepSeek-V3's strengths and weaknesses?
Relative to other ranked models, DeepSeek-V3 places best in coding, writing & preference, instruction following and lowest in long context, reasoning, math.
What is DeepSeek-V3 best at?
Its best category is coding, where it ranks 106th on Noometry.