DeepSeek V3
DeepSeekTEXT Open weight
- Provider
- DeepSeek
- Modality
- TEXT
- Parameters
- 671B
- Context window
- 128,000 tokens
- Weights
- Open
- Released
- —
DeepSeek V3 is a large, cost-efficient open-weight model with strong reasoning and coding ability.
It is a favourite for affordable, high-quality inference, whether self-hosted or via DeepSeek's very low-cost API.
Best for
- High-quality answers at low cost
- Reasoning and coding tasks
- Self-hosting where you have strong hardware
Strengths & how it compares
- Closest open model to top closed models on many tasks.
- Extremely cheap API pricing.
- Vs Llama 3.3 70B: stronger on reasoning; needs more hardware to self-host (large MoE).
Benchmarks
Representative public scores (approximate, higher is better) for relative comparison. Check the provider for the latest official results.
MMLU (knowledge)88
GPQA (reasoning)59
HumanEval (coding)89
How to access
Open weights; very low-cost API.
Access DeepSeek V3