Glossary term
Glossary term
Agentic Systems
Open-source code LLM family (1B to 33B) pre-trained on 2T tokens from 87 programming languages, achieving state-of-the-art results among open-source code models.
DeepSeek-Coder-33B achieves 56.1% on HumanEval and outperforms CodeLlama-34B by 7.9% on Python HumanEval, used as the default open-source coding model by enterprises needing self-hosted code-completion without proprietary API costs.
DeepSeek-Coder-V2 (2024, MoE architecture) achieves 90.2% on HumanEval, matching GPT-4o on coding benchmarks while being open-weight, and is the first open-source model to score above 10% on SWE-bench.
DeepSeek-Coder is integrated into Continue.dev, Cursor, and Tabby (open-source Copilot alternative) as the preferred open-weight code model for teams self-hosting their coding assistant on-premises for IP protection.