Ring-1T: three-stage training for a 1-trillion-parameter MoE
A post published on October 22, 2025 describes Ring-1T, a reasoning MoE model with 1 trillion parameters and about 50 billion active per token, trained in three stages.
Source: Ring-1T combina MoE de um trilhão de parâmetros com treinamento RL em três fases (x.com). Text prepared with AI from this source.
What happened and what to do
The post, published on October 22, 2025, describes Ring-1T, a reasoning MoE model with 1 trillion parameters and about 50 billion active per token. Its training combines SFT with long-CoT, reasoning RL with verifiable rewards, and general RLHF. The text also names IcePop, C3PO++, and ASystem as components of approaches to scaling this training.
For a company, the case highlights the importance of structuring data, evaluation criteria, and infrastructure before expanding AI systems. A company could implement evaluation using verifiable tasks and rewards, quality and cost monitoring, and a controlled pipeline for testing models on private data. The right approach depends on the company’s processes and security requirements.
How the consultancy can help
Wendelmaques can diagnose the opportunity and its data, security, and infrastructure requirements, propose a scoped implementation for evaluation and monitoring, and support its operation.
Next step
Send Wendelmaques a short description of your case to discuss it and receive a scoped proposal.
Let's assess your project
I reply with scope and a proposal.