Ring-2.6-1T
InclusionAI's Ring-2.6-1T is a 1 trillion parameter Mixture-of-Experts model with ~32B activated parameters per token, released by Ant Group's AI research arm. Targets frontier reasoning and code at MoE serving cost. Apache 2.0 license. Practical local deployment requires a multi-GPU workstation (8×H100 or 8×A100 80GB) or a high-memory Mac cluster; the activated-param count keeps token-generation cost in workstation territory.
Our verdict
The first open-weight 1T MoE that ships with a permissive license. The activated-param count (~32B/token) means inference is workstation-class, not cluster-class. The catch: getting the weights to the GPU is the workstation-class part — 8×H100 minimum. For solo operators this is "use the hosted API and migrate later if it matters." For labs with workstation racks, this is the first credible open Opus/Gemini Ultra-tier model.
Overview
InclusionAI's Ring-2.6-1T is a 1 trillion parameter Mixture-of-Experts model with ~32B activated parameters per token, released by Ant Group's AI research arm. Targets frontier reasoning and code at MoE serving cost. Apache 2.0 license. Practical local deployment requires a multi-GPU workstation (8×H100 or 8×A100 80GB) or a high-memory Mac cluster; the activated-param count keeps token-generation cost in workstation territory.
How to run it
Ring-2.6-1T is frontier-class — local deployment is workstation or cluster only. The realistic local path is 8×H100 80GB (640GB total HBM) via vLLM with tensor parallelism. On Apple Silicon, a 4-node M3 Ultra Mac Studio cluster (768GB unified) can run Q4 via MLX-distributed. Most users will hit the model through Together / DeepInfra hosted endpoints rather than self-hosting.
Strengths
Weaknesses
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of Ring-2.6-1T.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
Can I use Ring-2.6-1T commercially?
What's the context length of Ring-2.6-1T?
Source: huggingface.co/inclusionAI/Ring-2.6-1T
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify Ring-2.6-1T runs on your specific hardware before committing money.