Cast AI adds MiniMax M3 to Kimchi Coding as default model

Cast AI has added MiniMax M3 to its Kimchi Coding platform, making Kimchi the first autonomous coding agent to offer access to the model.

The move puts MiniMax M3 at the centre of Kimchi's model selection system as the default builder model for coding tasks. Kimchi is designed to route work across different AI models rather than rely on a single system for every job.

That approach reflects a wider shift in corporate AI use, as businesses weigh cost, control and performance when deciding how to deploy models. Companies are increasingly mixing open-weight models with commercial frontier systems, reserving more expensive tools for the most demanding work.

Kimchi Coding is a multi-model coding agent that assigns tasks based on complexity and cost. Its orchestration layer scores generated code, runs feedback loops and tracks token use in real time, while allowing spending limits to be set for individual users and organisations.

MiniMax M3 is an open-weight model that, according to Cast AI, performed strongly on SWE-bench Pro, a benchmark based on real GitHub issues. The company said the model scored 59% on the test, ahead of several commercial rivals included in the same evaluation.

Cast AI also highlighted the model's architecture, saying it reduces compute use for long-context workloads and increases decoding speed. Those features matter for software development teams running long coding sessions or working across large codebases, where processing costs can quickly rise, it argued.

Laurent Gil, President and Co-Founder of Cast AI, linked the launch to customer demand for lower-cost AI development tools with more control over data and deployment.

"We built Kimchi to give every developer frontier-quality AI coding without frontier-sized bills or data risk, thanks to our built-in token-optimized model orchestrator," said Laurent Gil, President and Co-Founder of Cast AI. "MiniMax M3 is the model Kimchi was designed for: open weights, frontier performance, and economics that work at scale."

Internal shadow-mode tests compared Kimchi against a baseline using only commercial AI models. In those tests, Cast AI said Kimchi reduced costs by 2.5 times while matching or exceeding the baseline on specification matching and test pass rates.

Deployment options

MiniMax M3 is being introduced through Kimchi's early access programme in phases. Organisations can run the service as a serverless deployment on Cast AI's inference clusters or install it in their own environments across AWS, GCP, Azure or on-premises systems.

The product also supports air-gapped deployments for customers that want stricter control over where code and model interactions take place. That is likely to appeal to businesses in regulated sectors, where concerns over data handling and model access have slowed adoption of public AI tools.

MiniMax also framed the partnership around demand for open-weight systems in business settings. Its marketing leadership said Kimchi offered a route for teams seeking greater control over where AI tools run.

"Cast AI's Kimchi Coding is an ideal partner to bring M3 to teams that demand both performance and sovereignty," said Leanna Ren, VP Global Marketing at MiniMax. "M3 was built to push the boundaries of open-weight models. Kimchi's model orchestrator and data sovereignty controls make it the natural home for M3 in the enterprise. Together, we're giving developer teams frontier-quality AI coding inside their own environment, at a fraction of the cost of commercial models alone."

Cast AI develops automation software for cloud-native and AI workloads and owns Kimchi Coding. It said it was valued at more than USD $1 billion after a strategic investment and counts BMW, Cisco, FICO, Hugging Face, Samsung and Swisscom among its customers.