Don't Sleep on OpenSource LLMs
Free beats $20 or $200 per month. Further, the fact that users may download and customize opensource models is attractive to many who perform deep research and coding. If privacy is a concern, you can’t beat a locally hosted opensource LLM, something one can’t do with proprietary models such as GPT, Claude, Grok and Gemini. The below is an excerpt from our recent report: Comparative Application Domains of Leading LLMs
Open-Source LLMs (DeepSeek et al.) play a major role in coding, especially in organizations or communities that prefer not to depend on proprietary AI. DeepSeek was specifically built for coding excellence, and its results bear that out: it outperforms Llama 2 70B on reasoning, coding, and math benchmarks, and in coding benchmarks like HumanEval it scores impressively high (Pass@1 of ~74% for DeepSeek-67B chat). It even demonstrated capabilities by achieving top marks on a national high school math exam. The significance of open models is that developers can self-host them and fine-tune for their codebase. For example, a company might use an open 7B or 13B model fine-tuned on its internal code repositories to assist developers with very domain-specific code suggestions – something ChatGPT cannot do out-of-the-box due to data privacy. Indeed, scientists and engineers globally downloaded DeepSeek over 10 million times from HuggingFace, indicating huge interest. Open models are also behind some popular coding agent projects (for instance, open-source “AutoGPT”-like agents that write and execute code to solve tasks often use a local LLM as the brain). Additionally, consider regions like China where access to ChatGPT is restricted; open models like DeepSeek (developed by a Chinese startup and available for anyone to download as “open weight”) fill the gap to empower local developers. A Nature news article from Jan 2025 highlighted that DeepSeek-R1 performs on par with OpenAI’s models in reasoning tasks and is “thrilling scientists as an affordable and open rival” for tasks like coding. It’s clear that open-source LLMs have become the backbone of many coding and devops workflows, especially where transparency and control are needed. The trade-off is that using them requires more setup and they might not have a slick chat interface unless one uses a community UI. But for those who invest the effort, open LLMs can be customized into extremely potent coding assistants.