AI startup Recursive has introduced Flint, a coding model designed to run locally on laptops. Backed by Nvidia and GV, the company is betting on its local-first approach to challenge cloud-based coding tools. Investors are tracking how this $4.65 billion entity balances compute costs with its goal of fully autonomous model development.
Recursive, the AI startup led by former Salesforce chief scientist Richard Socher, has launched Flint, an AI model for coding that runs directly on consumer laptops. This move shifts the focus from traditional cloud-heavy processing to local computing, potentially reducing long-term dependence on server-side infrastructure for software developers. By keeping the processing on the machine, the company is attempting to offer a faster and more private alternative to standard usage-fee models.
Local-First Coding Strategy
Unlike cloud-based tools that typically charge developers based on usage, Flint is designed to function offline. It uses a structure called mixture-of-experts, which allows the model to stay efficient by only using a small part of its total capacity for specific tasks. This design is critical for maintaining speed on local hardware. Early testing shows the model can achieve high processing speeds on hardware like the MacBook Pro with the M4 Pro chip. However, the true test will be whether local-first models can maintain the same accuracy and complexity handling as massive enterprise cloud platforms.
Efficiency Through Automation
The company claims it built Flint using its own automated system, the Eureka Machine. This internal system handled the bulk of the model’s lifecycle, including training and optimization. By using this automated research approach, Recursive reports a 71% reduction in model size while keeping performance high. This strategy allows the company to scale its research and development without needing to hire a massive team. While this automated approach to model-building is innovative, it remains a technical gamble, as investors will be watching to see if self-improving AI systems can consistently meet the high quality and safety standards required for commercial software development.
Capital and Competitive Landscape
Recursive has gained significant attention with a $4.65 billion valuation following $650 million in funding. The company counts high-profile investors like Nvidia and GV among its backers. However, the business is capital-intensive. Recursive recently locked in a $410 million compute investment deal with Amazon Web Services. This highlights the high cost of building and maintaining advanced AI models. With the release of Flint under an open-source Apache 2.0 license, the company is attempting to gain rapid adoption within the developer community. The long-term success of this approach depends on whether Recursive can sustain its research pace while managing these heavy capital costs and competing against well-funded enterprise coding assistants.