JetBrains launches Mellum: Ultra-fast LLM for developers

What happened?

JetBrains, the Czech company known for its integrated development environments (IDEs) such as IntelliJ IDEA, PyCharm, and WebStorm, has introduced Mellum, a large language model (LLM) specifically optimized for low latency and high performance in development workflows. The tool was initially spotted on Product Hunt, where it is described as 'Fast LLMs for low-latency and high-performance workflows'. Although JetBrains has not yet made a detailed official announcement, its appearance on Product Hunt suggests an imminent launch or open beta. This move marks a significant milestone, as it is the first time JetBrains has developed its own model instead of integrating third-party solutions like OpenAI or Anthropic.

Why is it important?

Mellum represents JetBrains' direct entry into the competitive market of AI assistants for development. Until now, the company had integrated AI features into its IDEs, such as the JetBrains AI Assistant, which relies on external models. With Mellum, JetBrains aims to offer an alternative to GitHub Copilot (based on GPT-4 and Codex), Amazon CodeWhisperer, and other competitors like Tabnine and Codeium. The promise of low latency is crucial for real-time user experience, especially in autocompletion and code suggestions. According to industry data, latency above 200 ms in code suggestions can disrupt a developer's workflow, reducing productivity by up to 30%. JetBrains, with its long track record in high-performance tools, seeks to capitalize on its experience in IDE optimization to deliver an LLM that runs efficiently, possibly even on local hardware.

Consequences for developers and companies

For developers, Mellum could translate into faster responses and deeper integration with JetBrains IDEs. Companies using the JetBrains ecosystem could benefit from a tool that does not require switching platforms, reducing migration and learning costs. However, the lack of detailed information on pricing, availability, and deployment models (local vs. cloud) creates uncertainty. A key aspect is code privacy: many companies avoid cloud solutions for fear of leaks. If JetBrains offers a local version of Mellum, it could attract sectors like banking, defense, and healthcare, where confidentiality is critical. On the other hand, the company might adopt a hybrid model, similar to what GitHub Copilot does with its option not to send code to servers, although this would limit the quality of suggestions.

Analysis and context

Historically, JetBrains has bet on local, high-performance tools. Its code search engine, for example, is designed to run on the user's machine without relying on the cloud. Mellum seems aligned with that philosophy, although LLMs typically require significant server resources. The company could offer an optimized local version through techniques like quantization or distillation, or a cloud service with low latency thanks to geographically distributed servers. The comparison with GitHub Copilot is inevitable: Copilot integrates into multiple IDEs, but Mellum could have an advantage in JetBrains IDEs due to its native and deep integration, with access to features like refactoring, debugging, and static code analysis. Additionally, code privacy is a key factor: JetBrains could promise that code is not sent to external servers, attracting security-conscious companies. In contrast, GitHub Copilot has faced criticism for potential open-source license violations, a problem JetBrains could avoid by training its model on data from its own repository or permissive licenses.

Another relevant aspect is the AI assistant market, which has experienced explosive growth. According to a Grand View Research report, the global AI in software development market is expected to reach $1.2 billion by 2028, with a compound annual growth rate of 28%. JetBrains, with an installed base of over 10 million developers, has a unique opportunity to capture a significant share. However, competition is fierce: GitHub Copilot already has over 1.3 million paid subscribers, and Amazon CodeWhisperer is free for individual use. Mellum will need to differentiate itself not only by performance but also by price and features.

What readers should know

• Mellum is an LLM for development, not yet generally available. It was spotted on Product Hunt on [date], but without official confirmation from JetBrains.
• Its main promised advantage is low latency, which could improve the real-time autocomplete experience.
• It is designed to integrate with JetBrains IDEs, possibly natively, leveraging the existing ecosystem.
• No pricing or release date has been confirmed. Speculation suggests it could be free for JetBrains IDE users or have a subscription model.
• It could directly compete with GitHub Copilot, Amazon CodeWhisperer, Tabnine, and Codeium, but with potential advantages in integration and privacy.
• JetBrains has a history of local, high-performance tools, suggesting Mellum could offer an on-premise option.

Conclusion

JetBrains' Mellum is a strategic move to consolidate its development ecosystem with its own AI. The promise of low latency and high performance is attractive, but key will be seeing results in independent benchmarks and community adoption. The company will need to balance innovation with transparency regarding privacy, pricing, and availability. If it can integrate Mellum seamlessly into its IDEs and maintain the low-latency promise, it could become a serious competitor to GitHub Copilot. TheVortiq will closely follow this launch and provide updates as more information becomes available.