It recently seemed that true AI agents and breakthrough tasks were exclusively the domain of closed commercial models. Companies like OpenAI and Anthropic likely grew accustomed to the idea that progress was solely their purview. However, recent weeks have clearly demonstrated that open-source Large Language Models (LLMs) have crossed a threshold beyond which they cannot be ignored. Models such as GLM-5 from z.ai and MiniMax M2.7 are delivering results comparable to top-tier closed-source counterparts in key tasks. Whether it's working with files, utilizing third-party tools, or simply following instructions, open-source has caught up. For businesses already familiar with the pricing appetites and latency of proprietary solutions, this development represents a significant shake-up.
Stripping away the marketing fluff, the main advantage of open-source lies in pure economics and speed. Closed-source models can be eight to ten times more expensive under substantial loads. Consider this example: processing 10 million tokens per day on Opus 4.6 costs $250. The same task handled by MiniMax M2.7 comes in at $12. This difference of $87,000 annually is a compelling argument to stop feeding the giants. Furthermore, closed-source models are often sluggish for interactive products where every millisecond is critical. Open-source models, thanks to optimization and smaller dimensions, demonstrate remarkable performance on specialized infrastructure from providers like Groq, Fireworks, or Baseten. GLM-5 on Baseten achieves a latency of 0.65 seconds (70 tokens/sec), while Opus 4.6 lags at 2.56 seconds. This is a chasm that is difficult to overlook.
These conclusions are not baseless; they are supported by test results conducted using the developments from Deep Agents. Researchers evaluated seven categories of tasks, including the reliability of tool invocation, instruction following, and file handling. Both execution accuracy and speed were assessed. GLM-5 and MiniMax M2.7 have shown that open-source models are production-ready. They provide sufficient predictability and stability for implementation into real-world workflows. These are no longer just toys for enthusiasts.
Certainly, transitioning to the open-source world is not a walk in the park. Businesses should soberly weigh not only the promised benefits but also the potential risks. Data security, the necessity of self-support, integration complexities, and the absence of guaranteed technical assistance all require additional expertise and resources. The flexibility of open-source is a double-edged sword, and readiness for this must be considered.
This means that open LLMs have ceased to be a niche solution for geeks. They are now a tangible tool for reducing costs and accelerating interactive AI services. Companies actively implementing AI gain an opportunity to compete with major players on more even terms by leveraging accessible and powerful technologies.