Ring-2.6-1T
Editorial notes
Lanzado 8 mayo 2026 por inclusionAI (lab de Ant Group). Flagship de razonamiento. MoE 1T total / 63B activos. 128K → 262K context (YaRN). Open-weight MIT. Scores self-reported del model card oficial huggingface.co/inclusionAI/Ring-2.6-1T (config xhigh): GPQA-Diamond 88.27 (2do mas alto del catalogo tras Mythos), SWE-bench-Verified 74, ARC-AGI-2 66.18, Tau2-Bench Telecom 95.32. AIME 2026: 95.83 (Ant publica AIME 2026, NO 2025 — por eso no entra a la columna AIME-2025 del core-set). Artificial Analysis Intelligence Index 31 (tercero). MMLU-Pro/Terminal-Bench/LiveCodeBench/HLE/SWE-Pro no publicados. NOTA: con solo 2 benchmarks del core-set (GPQA-D + SWE-V; AIME-2026 y ARC-AGI-2 quedan fuera del core-set) no alcanza el gate de elegibilidad de 3 — ver decision pendiente sobre AIME-2026.
Spec sheet
- Company
- Ant Group
- Country
- CN
- Type
- reasoning
- Release
- 2026-05
- Context
- 262K tokens
- Params total
- 1.0T
- Params active (MoE)
- 63B
- License
- mit
- Slug
- ring-2-6-1t
Benchmarks (3)
Cite this model
BibTeX · APA
BibTeX
@misc{frontier-ring-2-6-1t,
title = {Ring-2.6-1T},
author = {{Ant Group}},
year = {2026},
note = {Frontier Benchmarks AI atlas. Accessed 2026-06-17},
url = {https://frontierbenchmarks.com/models/ring-2-6-1t}
} APA
Ant Group (2026). Ring-2.6-1T [Large language model]. Frontier Benchmarks AI. Retrieved 2026-06-17, from https://frontierbenchmarks.com/models/ring-2-6-1t
Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.