MiniMax M3 Comes to Kimchi: The first Coding Platform Worldwide to Launch the Frontier Open-Weight Model
MiniMax M3, the open-weight model that topped SWE-Bench Pro with a 59% score, is available through Kimchi Coding. Developers can begin using M3 immediately inside their own cloud.
MIAMI — June 12, 2026 — Cast AI today announced that MiniMax M3, the open-weight model that topped SWE-Bench Pro with a 59% score (according to MiniMax), is available through its Kimchi Coding platform. Developers who join the Kimchi Early Access program at kimchi.dev can begin using M3 immediately - deployed inside their own AWS, GCP, Azure, or on-prem environment.
M3 is the first open-weight model to combine frontier coding ability, a one-million-token context window, and native multimodal understanding of text, image, and video in a single architecture. Its Sparse Attention design delivers 15.6× faster decoding, cutting inference costs to roughly 1/20th of previous levels. For enterprises that need top-tier coding performance without sending proprietary code to a third-party API, M3 inside Kimchi closes the gap between open-weight and commercial models.
Kimchi Coding
Kimchi Coding includes an autonomous multi-model harness that delivers best-in-class coding performance at a 12x lower cost. Its model orchestration engine autonomously routes each task to the best-fit model based on complexity and cost, reserving frontier models for the hardest evaluations while other open-weight models handle the bulk of the work. Budget governance with built-in token optimization enforces hard spend caps from individual API keys to entire organizations, with automatic termination of runaway agentic loops. A real-time FinOps dashboard tracks every token, user, and model with cost attribution per developer, team, and project.
In continuous shadow-mode evaluations against a commercial-models-only baseline, Kimchi Coding delivers 12x lower cost and reduces latency by 12% while matching or exceeding quality - spec match of 9.1 versus 9.0 and test pass rates of 97% versus 96%. Kimchi runs serverless on Cast AI inference cloud and can also be deployed as a sovereign platform that keeps models inside the customer's own cloud with full air-gap support and compliance readiness for SOC 2, GDPR, ISO 27001, and PCI-DSS.
"We built Kimchi to give every developer frontier-quality AI coding without frontier-sized bills or data risk," 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."
"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: frontier coding performance, a one-million-token context window, and native multimodal understanding in a single architecture. 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".
Join the early access program at kimchi.dev.
About Kimchi
Kimchi, by the creators of Cast AI
Cast AI is the leading automation platform for cloud-native and AI infrastructure. The company is backed by Softbank, G2 Venture Partners, Hedosophia, Creandum, Vintage Investment Partners, Cota Capital, Aglae Ventures, and Uncorrelated Ventures. It achieved unicorn status in January 2026 with a strategic investment from Shinsegae Group and Kakao Corp., valuing the company at over $1 billion. Cast AI is trusted by BMW, Cisco, FICO, Hugging Face, and Swisscom to keep mission-critical applications reliable and performant at scale. Learn more at cast.ai.