Ling 2.6 1T

Overview

Ant Group's Ling 2.6 1T: MIT-licensed MoE with 50B active params, hybrid MLA + Linear Attention, and 256k context. Top open non-reasoning model with an Intelligence Index of 34.

When to pick this model

• Agentic workloads needing mature tool calling at frontier scale
• Long-context analysis up to 256k tokens
• MIT-licensed datacenter deployments
• Non-reasoning workloads where speed beats deliberation
• Replacing closed flagships with open weights

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	580 GB
Q5_K_M	710 GB
Q8_0	1070 GB
FP16 (no quantization)	2000 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Published benchmark scores

Benchmark	Score
AA Intelligence Index	34

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

Strengths

• Permissive MIT license
• Top open non-reasoning Intelligence Index (34)
• 256k context window
• Efficient hybrid MLA + Linear Attention
• Mature agentic tool calling, compatible with Qwen2.5 parsers

Limitations

• Around 600 GB VRAM in Q4 — datacenter required
• Hugging Face weights only — no Ollama tag
• Not a reasoning model; pick DeepSeek V4 for deliberation

Architecture & training

Architecture: BailingMoeV2.5 · MoE 1T total / 50B active · 256 experts top-8 + 1 shared · 80 layers · hybrid MLA + Linear Attention · 256k ctx

Training: Ling 2.6 family (Ant Group). Contextual Process Redundancy Suppression and 'Fast Thinking' strategy to reduce token overhead. Qwen2.5-compatible tool-call parser.

Quick start

# HuggingFace : inclusionAI/Ling-2.6-1T

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.