MiniMax is raising $2 billion while its CEO has pledged part of his equity to support future open-source projects, strengthening the company’s investment in open-source AI models and infrastructure.
Shanghai-based MiniMax, the developer of the open-source MiniMax family of large language models (LLMs), is raising $2 billion in fresh funding while reinforcing its commitment to open-source AI. According to reports, CEO Yan Junjie also plans to donate 5% of his personal shareholding to finance employee incentives and future open-source projects.
More than half of the funding is expected to come from newly issued shares, with the remainder raised through convertible bonds. MiniMax also reportedly plans a further $6.5 billion zero-coupon convertible bond issuance, allowing investors to acquire 19.4 million shares at a price 12.6% above the company’s Thursday closing price. The bonds mature in 2027.
The company’s latest open-source flagship, MiniMax-M3, launched in June with 427 billion parameters and a one million-token context window. MiniMax says the model delivers prefill performance nine times faster and decode performance 15 times faster than its previous flagship.
The gains are attributed to MiniMax Sparse Attention (MSA), which builds on FlashAttention to reduce data movement between SRAM and HBM during inference. MSA also incorporates block-sparse prefill for faster processing of large prompts and quantisation to lower memory requirements.
Beyond LLMs, MiniMax maintains the open-source VTL family of visual tokenisers for multimodal AI. The company generates revenue through hosted AI models and consumer multimedia generation applications. Earlier this year, it raised about $619 million in its Hong Kong IPO, while its shares fell 9.8% following reports of the latest fundraising.
