DeepSeek V3 Open-Sourced, Performance Rivals GPT-4
DeepSeek announced full open-source release of V3 model, excelling across multiple benchmarks.
DeepSeek announced full open-source release of V3 weights and inference code, reaching GPT-4-level performance on MMLU, HumanEval, and other major benchmarks, drawing strong attention from the open-source community.
The team said the strategy aims to advance global AI research and collaboration while offering commercial licensing options.
Score · 0–100
- 72
Twitter/X Heat
Weight 22%
- 81
GitHub Growth
Weight 20%
- 76
Hacker News
Weight 15%
- 73
Google Trends
Weight 18%
- 94
Official Release Weight
Weight 25%
Impact Fields
Why It Matters
High-performance open weights challenge closed-source pricing and heat up compute and inference cost debates.
- ›GPT-4-class performance self-hostable
- ›Commercial license options available
- ›Open-source community buzz surging
Industry Impact
SMBs can build self-hosted inference stacks and reduce reliance on single API vendors.
Competitive Shift
Chinese AI moves beyond API wrappers toward a differentiated path vs. OpenAI.
For Developers
Watch quantization, vLLM deployment, and vertical fine-tuning—costs can drop an order of magnitude.