MiniMax and Together AI will bring open-model economics to London
The two AI companies will host Rio Shen, Max Ryabinin and Sarung Tripathi at a September 16th event on the production economics of open models.
By RuntimeWire Staff · Published
Primary source: MiniMax
Why it matters
MiniMax and Together AI are making a production-cost case for open-weight models. Buyers still have to account for routing, reliability, licensing and infrastructure before lower token prices translate into a smaller bill.
MiniMax, founded by Yan Junjie, and Together AI, co-founded by Vipul Ved Prakash (@vipulved), will put their shared argument for cheaper, open-weight AI in front of London engineers on September 16th.
The event, titled "Open by Design: The Economics of AI in Production", is scheduled for 10 York Road in London. The event listing frames the discussion around cost savings, model selection and routing, and moving open models into production while preserving performance and reliability.
The scheduled speakers are Rio Shen, MiniMax's general manager for Europe, the Middle East and Africa; Max Ryabinin, Together AI's vice president of model shaping; and Sarung Tripathi, Together AI's vice president of customer experience.
Those subjects give MiniMax and Together AI a stage for discussing work they have already done together. Together AI has published an engineering account of serving MiniMax M3. The London program carries that engineering relationship into a discussion aimed at production teams.
Yan and Prakash arrive at the same cost argument
Yan founded MiniMax in early 2022 after spending more than six years at SenseTime, where he became a vice president and deputy head of its research institute. MiniMax's management biography says Yan studied mathematics at Southeast University, earned a doctorate in artificial intelligence from the Institute of Automation at the Chinese Academy of Sciences and completed postdoctoral research at Tsinghua University.
That research background now feeds a commercial model portfolio spanning language, video, speech and music. MiniMax's language lineup includes M3, a multimodal coding model with a context window of up to 1 million tokens.
Prakash has built Together AI around the infrastructure side of the same equation. In announcing Together AI's $800 million Series C, he wrote that he and his co-founders started Together AI four years earlier because they believed generative AI should be "open and abundantly available" rather than controlled by a small number of vendors.
Together AI provides compute, inference, fine-tuning and model-shaping services for models from providers including MiniMax. Lower model costs can bring more workloads into production, while each production workload creates demand for the infrastructure needed to run it.
MiniMax benefits from broader distribution. Together AI gives developers another way to run M3 without taking on the hardware and serving work involved in deploying a large model. The London event places MiniMax's regional sales leadership alongside the engineers and customer operators responsible for making that deployment case credible.
The cost argument needs production data
A low token price remains only one variable. A cheaper model can lose its advantage if an application needs repeated calls, longer outputs, more retries or additional verification to reach the required result. Routing adds another layer: sending routine work to a smaller model and difficult requests to a larger one can cut spending, provided the router makes that distinction accurately.
Together AI says some customers using open models have achieved costs six to 20 times lower, including a claimed sixfold reduction for customer-service software provider Decagon. Those are company-reported figures, and Together AI has not published enough workload-level data in the cited announcement to generalize them across applications.
The September 16th session gives Shen, Ryabinin and Tripathi a chance to replace broad multiples with the details production teams need: which workloads moved, what quality threshold was held constant, how routing changed token consumption and what engineering expenses were included.
"Open" still comes with terms
MiniMax and Together AI use the language of open models because M3's weights can be downloaded and deployed outside MiniMax's API. That freedom has boundaries.
The MiniMax Community License requires commercial products using M3 to display a "Built with MiniMax M3" notice. Products or services generating more than $20 million in annual revenue need prior written authorization from MiniMax for commercial use, while smaller commercial users must send the company a one-time notice.
M3 is an open-weight model distributed under a vendor license. It does not carry the unrestricted terms commonly associated with open-source software. The distinction belongs in any production cost calculation because licensing, compliance and vendor approval can matter alongside GPU hours and token charges.
The September 16th event moves the argument from model claims to production budgets. MiniMax wants its models selected, and Together AI wants to supply the infrastructure. London engineers will be looking for the calculation that proves both choices hold up after the first invoice.