DeepSeek
DeepSeek V3
deepseek-v3Context
128K
Input / 1M tokens
$0.1400
Output / 1M tokens
$0.2800
Cached input / 1M tokens
— (bills at input price)
Cache write / 1M tokens
— (bills at input price)
Available in both lanes: guarded and community. Prices above are for the community route (deepseek-v3); see Ways to call for each route.
About this model
DeepSeek V3
DeepSeek V3 is a 671B-parameter Mixture-of-Experts model that activates 37B parameters per token. Designed for general-purpose chat, document analysis, and structured output, it punches well above its cost class on coding, reasoning, and long-context comprehension. The MoE architecture means inference is roughly as fast as a dense 37B model while quality matches dense 70B+ flagships.
Key facts
- Maker
- DeepSeek
- Context window
- 128K tokens
- Max output
- 4K tokens
- Lane
- Guarded and Community (community routes need account opt-in and the provider may train on data; see Ways to call below)
- Modality
- Text
- Input price
- $0.1400 per 1M tokens (community route)
- Output price
- $0.2800 per 1M tokens (community route)
- Processing region(s)
- cn
- Route
- Via SiliconFlow's API
- Cheapest plan
- Available from the Plus Pass plan
- Price comparison
- Priced below 6 of 15 plus-band community models on Anoman (blended price: 1 part input to 3 parts output).
Released
2024-12
Training cutoff
2024-07
Parameters
MoE 671B (37B active)
Ways to call
Ways to call DeepSeek V3
Each model ID below is callable through the Anoman AI API. Price and processing region depend on the route.
| Model ID | Route | Input / output per 1M tokens | Region(s) | Lane | Context |
|---|---|---|---|---|---|
| deepseek-v3 | Via SiliconFlow's API | $0.1400 / $0.2800 | cn | Community (opt-in, may train) | 128K |
| openrouter/deepseek-v3 | Via OpenRouter | $0.2900 / $1.14 | us | Guarded | 164K |
| together/deepseek-v3 | Via Together AI's API | $1.25 / $1.25 | us | Guarded | 66K |
Best for
Use cases
- Cost-sensitive production chat
- Long-context document analysis (128K window)
- Code generation + review
- Structured-output extraction (JSON, function calls)
- Multilingual workloads (Chinese, English, code)
Strengths
What it does well
- Top-tier coding ability for the price (HumanEval 89%)
- Strong on Chinese-language benchmarks (CMMLU 88.8)
- Native function calling + JSON mode
- 128K input + 8K output — fits most agent loops
- Mixture-of-Experts keeps inference cheap at scale
Limitations
Know the trade-offs
- Knowledge cutoff is mid-2024 — agents may need search augmentation for fresh facts
- Reasoning depth trails specialist reasoners (use DeepSeek R1 for math-heavy work)
- Outputs longer responses than GPT-4o by default — set max_tokens explicitly for cost control
Benchmarks
Published scores
Scores from official model cards and public leaderboards. Higher is better unless noted.
| Benchmark | Score | Measures |
|---|---|---|
| MMLU | 87.1 | General knowledge across 57 subjects |
| MMLU-Pro | 75.9 | Harder MMLU variant, 10 options per question |
| HumanEval | 89.0 | Python code generation, pass@1 |
| MATH-500 | 89.3 | Competition mathematics |
| DROP | 90.1 | Reading comprehension + numeric reasoning |
Code example
Chat completion with DeepSeek V3
from openai import OpenAI
client = OpenAI(
base_url="https://api.anoman.io/v1",
api_key="anm-sk-..."
)
response = client.chat.completions.create(
model="deepseek-v3",
messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)Alternatives
Similar models in our catalog
Same maker first, then the closest price to DeepSeek V3 in the Anoman catalog.