{"id":582041,"date":"2026-04-12T01:37:16","date_gmt":"2026-04-12T01:37:16","guid":{"rendered":"https:\/\/Blockchain.News\/news\/minimax-m27-230b-parameter-ai-model-nvidia-infrastructure"},"modified":"2026-04-12T01:37:16","modified_gmt":"2026-04-12T01:37:16","slug":"minimax-m2-7-brings-230b-parameter-ai-model-to-nvidia-infrastructure","status":"publish","type":"post","link":"https:\/\/e-bitco.in\/index.php\/2026\/04\/12\/minimax-m2-7-brings-230b-parameter-ai-model-to-nvidia-infrastructure\/","title":{"rendered":"MiniMax M2.7 Brings 230B-Parameter AI Model to NVIDIA Infrastructure"},"content":{"rendered":"
\n

Ted Hisokawa<\/a> Apr 12, 2026 01:37<\/span> <\/p>\n

MiniMax releases M2.7, a 230B-parameter mixture-of-experts model optimized for NVIDIA GPUs with up to 2.7x throughput gains on Blackwell hardware.<\/p>\n

\"MiniMax <\/a> <\/figure>\n

MiniMax has released M2.7, a 230-billion parameter open-weights AI model designed specifically for autonomous agent workflows, now available across NVIDIA’s inference ecosystem including the company’s latest Blackwell Ultra GPUs.<\/p>\n

The model represents a significant efficiency play in enterprise AI. Despite its massive 230B total parameters, M2.7 activates only 10B parameters per token\u2014a 4.3% activation rate achieved through mixture-of-experts (MoE) architecture with 256 local experts. This keeps inference costs manageable while maintaining the reasoning capacity of a much larger model.<\/p>\n

Performance Numbers on Blackwell<\/h2>\n

NVIDIA collaborated with open source communities to optimize M2.7 for production workloads. Two key optimizations\u2014a fused QK RMS Norm kernel and FP8 MoE integration from TensorRT-LLM\u2014delivered substantial throughput improvements on Blackwell Ultra GPUs.<\/p>\n

Testing with a 1K\/1K input\/output sequence length dataset showed vLLM achieved up to 2.5x throughput improvement, while SGLang hit 2.7x gains. Both optimizations were implemented within a single month, suggesting further performance headroom exists.<\/p>\n

Technical Architecture<\/h2>\n

M2.7 supports 200K input context length across 62 layers, using multi-head causal self-attention with Rotary Position Embeddings (RoPE). A top-k expert routing mechanism activates only 8 of the 256 experts for any given input, which is how the model maintains low inference costs despite its scale.<\/p>\n

The architecture targets coding challenges and complex agentic tasks\u2014workflows where AI systems need to plan, execute, and iterate autonomously rather than respond to single prompts.<\/p>\n

Deployment Options<\/h2>\n

Developers can access M2.7 through multiple channels. NVIDIA’s NemoClaw reference stack provides a one-click deployment for running autonomous agents with OpenShell runtime. The model is also available through NVIDIA NIM containerized microservices for on-premise, cloud, or hybrid deployments.<\/p>\n

For teams wanting to customize the model, NVIDIA’s NeMo AutoModel library supports fine-tuning with published recipes. Reinforcement learning workflows are available through NeMo RL with sample configurations for 8K and 16K sequence lengths.<\/p>\n

Free GPU-accelerated endpoints on build.nvidia.com allow testing before committing to infrastructure. The open weights are also available on Hugging Face for self-hosted deployments.<\/p>\n

The release positions MiniMax as a credible alternative to closed models from OpenAI and Anthropic for enterprises building autonomous AI systems, particularly those already invested in NVIDIA infrastructure.<\/p>\n

Image source: Shutterstock<\/i><\/span> Source<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Ted Hisokawa Apr 12, 2026 01:37 MiniMax releases M2.7, a 230B-parameter mixture-of-experts model optimized for NVIDIA GPUs with up to 2.7x throughput gains on Blackwell hardware. MiniMax has released M2.7, a 230-billion parameter open-weights AI model designed specifically for autonomous agent workflows, now available across NVIDIA’s inference ecosystem including the company’s latest Blackwell Ultra GPUs. […]<\/p>\n","protected":false},"author":1,"featured_media":582042,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[12],"tags":[20916,20460,2572,24617,25,2148],"class_list":{"0":"post-582041","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-blockchain","8":"tag-ai-infrastructure","9":"tag-enterprise-ai","10":"tag-machine-learning","11":"tag-minimax","12":"tag-news","13":"tag-nvidia"},"_links":{"self":[{"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/posts\/582041","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/comments?post=582041"}],"version-history":[{"count":0,"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/posts\/582041\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/media\/582042"}],"wp:attachment":[{"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/media?parent=582041"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/categories?post=582041"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/e-bitco.in\/index.php\/wp-json\/wp\/v2\/tags?post=582041"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}