Ling

Ling is a general-purpose large language model series independently developed and open-sourced by Ant Group. Built on the MoE (Mixture of Experts) architecture, it has been validated on domestic heterogeneous computing platforms, achieved breakthrough at the trillion-parameter scale, and continues to evolve with long context modeling and AI Agent collaborative reasoning capabilities.

Why Choose Ling?

Ling achieves key breakthroughs in 3 dimensions: inference efficiency, long context modeling, and training technology:

Inference Efficiency

Traditional Dense models require activating all parameters during inference, with computational overhead proportional to model size. Ling’s MoE architecture routes each token to activate only the most relevant expert subnetworks, compressing the actual computation to a minimal fraction while maintaining large-scale model capacity.

Long Context Support

Building on MoE’s efficient inference, the Ling series further advances long-sequence modeling.

Ling-3.0-flash: Native 256K context window, extendable up to 1M.
Ling-2.6-1T: Supports up to 1M context window natively; the official API currently exposes a 256K context window.
Long-range information retrieval without noticeable degradation: the model reliably retrieves information regardless of whether it appears at the beginning, middle, or end of the context.

Model Selection Guide

The following is a capability comparison of Ling models at different scales to help you evaluate and select:

Model	Positioning	Context	Parameters	Activated Parameters
Ling-3.0-flash	Cost-effective hybrid reasoning model with 5.1B activated parameters — high TPS for high-frequency tasks, peak inference speed up to 1000 tokens/s — fast, stable, and low cost	256K	124B	5.1B
Ling-3.0-tiny	Lightweight hybrid reasoning model with only 1.3B activated parameters — supports on-device deployment and local offline execution, low latency and low cost, built for real-time interaction and tool use	—	7.9B	1.3B
Ling-2.6-1T	Flagship model with 1M context length, covering the full chain from logical reasoning to task execution	1M	1T	63B

Selection Suggestions:

For most general-purpose scenarios, start with Ling-3.0-flash — it delivers higher efficiency under the same compute resources, handles complex tasks with greater ease, and significantly reduces cost.
Need to handle ultra-long documents, complex multi-hop reasoning, or Agent task chains? Choose Ling-2.6-1T, which leverages the innovative MLA and Linear Attention Hybrid architecture to bypass cumbersome “slow thinking” and achieve efficient inference through a “fast thinking” mechanism, reaching results with minimal token overhead and dramatically reducing output cost.
For online service scenarios sensitive to inference latency and throughput, Ling-3.0-flash’s low activated parameter count gives it superior TTFT (Time to First Token) and TPS (Tokens Per Second) performance.
Need on-device deployment, local offline execution, or extremely cost- and resource-constrained environments? Choose Ling-3.0-tiny — with only 1.3B activated parameters, it handles real-time interaction, lightweight code assistance, local knowledge base retrieval, and tool calling with ease.

Model Details

Next, let’s dive into the technical characteristics and Use Cases of each model:

Ling-3.0-flash

Ling-3.0-flash is the latest-generation cost-effective model in the Ling series, with 124B total parameters, 5.1B activated parameters, a native 256K context window extendable up to 1M.
Compared to the previous flash model, Ling-3.0-flash delivers significantly enhanced stability in long-horizon tasks, markedly higher tool calling accuracy, and substantially improved compatibility with common Harness environments. In coding tasks, Ling-3.0-flash features greatly optimized spatial understanding, capable of constructing physical scene grids and relative positions. Additionally, thanks to its hybrid Reasoning mode, task success rates across varying difficulty levels have all improved.

Advantages:

Cost-effective: lower activation, higher throughput — lower cost and higher efficiency per unit task, handling complex tasks with greater ease
Seamless switching between thinking and non-thinking modes — engage deep thinking for complex tasks, or disable it for high-volume production and low-latency scenarios
High TPS for high-frequency tasks: peak inference speed reaches 1000 tokens/s with TTFT < 100ms, enabling more reading, checking, and thinking within the same time unit for more iterations and more accurate results
Stable tool calling: under the RL strong-check mechanism, long-horizon tool calling stability is significantly improved with reliable instruction following
Native 256K context window, extendable up to 1M, with full tool calling and Agent capabilities

Use Cases:

Office automation, office collaboration, and Cowork
Human-AI collaboration: agile and efficient daily code development under human guidance
High-concurrency batch data processing and analysis scenarios, such as resume screening, document review, knowledge base organization, and data cleaning
Scenarios requiring high response speed, such as digital human live-streaming, streaming media interaction, and real-time translation

Ling-3.0-tiny

Ling-3.0-tiny is a lightweight hybrid reasoning model in the Ling series designed for edge deployment scenarios, with 7.9B total parameters and only 1.3B activated parameters. It supports on-device deployment and local offline execution, and will be open-sourced in the near future.
With minimal activated parameters, it delivers real-world task execution capability, combining mathematical reasoning, instruction following, and low hallucination, achieving balanced performance across scientific reasoning, code, real-world tasks, and long context.

Advantages:

Ultra-lightweight: only 1.3B activated parameters, deployable on edge devices and resource-constrained environments with low inference latency and minimal compute overhead
Comprehensive capability coverage: practical performance in mathematical reasoning, instruction following, code generation, and tool calling — “tiny” does not mean compromised capability
Native tool-use support, enabling lightweight automation scenarios such as mobile device and browser UI automation
Local offline execution support, ideal for privacy-sensitive private deployments where data must not leave the device

Use Cases:

On-device applications and local offline inference, such as local knowledge base retrieval and content generation
Lightweight code assistants, local web page translation, and other high-frequency low-latency tasks
Task agents in enterprise private deployments and model foundations for vertical scenarios
Tool calling and UI automation for real-time control scenarios

Ling-2.6-1T

Ling-2.6-1T is the latest-generation flagship model in the Ling series. It uses a Hybrid architecture combining MLA and Linear Attention, with approximately 1T total parameters and 63B activated parameters, supports a 1M ultra-long context, and delivers a strong balance between flagship-level intelligence and token efficiency.

Advantages:

Fast thinking replaces lengthy reasoning chains, preserving 1T parameter level capability at lower token cost.
Stronger coding and Agent capabilities, achieving open-source SOTA on execution-focused benchmarks such as AIME26 and SWE-bench Verified.
Highly compatible with mainstream Agent frameworks including Claude Code, OpenCode, and OpenClaw, covering multi-tool, multi-step, and multi-constraint scenarios.

Use Cases:

Multi-step tasks and agent tasks
Code completion, project delivery, and bug fixing
Data visualization projects such as slides and reports
Long-context knowledge management and automated workflows

Ling’s Evolution

Ling has undergone a complete technical evolution from computing platform validation to Agent capability breakthroughs:

Time	Version	Key Technical Breakthroughs
2025.03	Ling 1.0 Series	Validated the engineering feasibility of MoE large language models on non-high-end heterogeneous computing platforms (non-A100/H100), completing domestic computing platform adaptation
2025.10	Ling 2.0 Series	First breakthrough at trillion-parameter scale, introduced FP8 end-to-end training, significantly improving training efficiency and achieving cross-domain generalization
2026.02	Ling 2.5 Series	Achieved high-throughput decoding optimization, long context understanding capability reached industry-leading levels, initial Agent interaction and tool calling foundation
2026.04	Ling 2.6 Series	Completed the full reasoning chain from logical reasoning to task execution, significantly improved “intelligence-efficiency ratio” (task completion per unit compute), opening the era of cost-effective, scalable Agent applications
2026.07	Ling 3.0 Series	Transitioned from migration-based hybrid linear architecture to native hybrid linear design, improving the conversion efficiency of model capacity and computation with lower expert activation ratios, and introduced KDA (Kimi Delta Attention) to enhance memory update and information retention in long sequences

Technology Ecosystem

Based on the Ling foundation model, we have built a complete technology ecosystem covering the full pipeline from training to deployment:

High-Performance Operator Library: Open-sourced high-performance training and inference operator system, covering core components such as MoE routing and attention computation, supporting full-pipeline optimization from pre-training to online inference.
Vertical Domain Models: Domain-adapted models for specialized scenarios such as healthcare and finance, with superior performance on domain knowledge-intensive tasks.
Open Source Community: All research results open-sourced to the Inclusion AI community, continuously co-building the ecosystem with developers.

Quick Start

Create API Key: Obtain access credentials.
Make Your First Call: Complete your first API request in 5 minutes.
Explore More Capabilities: Learn more about advanced features such as inference optimization and long context.

Was this page helpful?