Chinese AI startup MiniMax is building a 2.7-trillion-parameter large language model, which it plans to open-source, according to The Information. The model, internally called M3 Pro, would be China's…
A developer built ignis, a from-scratch LLM inference engine in Rust with only two dependencies, to explore how compiler design principles apply to inference. The engine uses SSA IR, fusion passes, an…
On June 1, 2026, Shanghai-based AI lab MiniMax released M3, the first open-weight model combining frontier coding, a 1M-token context window, and native multimodal input. The model uses MiniMax Sparse…
MiniMax introduces M3, a sparse attention mechanism that efficiently processes up to a million tokens by selectively reading only relevant parts of the input, overcoming production failures of prior e…
A developer from Lekhai.app has published a guide on running AI models locally on a Mac, highlighting benefits such as privacy, speed, cost savings, and offline access. The guide covers requirements i…
Telnyx has added MiniMax's M3 model to its Inference platform, offering the first open-weight model combining coding, agent capabilities, and a 1M-token context window. Hosted on Telnyx's B300 GPU inf…
A new paper from MiniMax, MaxProof, trains a model to write competition-math proofs and achieves high scores on IMO and USAMO problems, but a cross-verification experiment reveals that among proofs th…
MiniMax released M3, an open-weight AI model with a 1 million token context window and native multimodal support, that outperformed frontier models in a 24-hour CUDA kernel optimization task by persis…
Together AI partnered with MiniMax to serve the new M3 model in production, achieving 81–125% throughput improvements through engineering optimizations including a KV-Block-Major sparse attention kern…
MiniMax M3, the company's first model with a 1M-token context window and native multimodality, is now available on Vercel AI Gateway. The model, built around MiniMax Sparse Attention, improves softwar…
Shanghai-based AI firm MiniMax has teased its next-generation M3 model, claiming a 15.6x faster decoding speed and 9.7x faster prefill speed over its M2 model when processing 1M-token contexts, driven…
A developer built a local-first movie recommendation system using a Corrective-RAG pipeline that runs entirely on Ollama. The system employs query expansion at ingestion time rather than query time, g…