🤗 Hugging Face | 🤖 ModelScope | 🐙 Experience Now
Ling-1T is the first flagship non-thinking model in the Ling 2.0 series, featuring 1 trillion total parameters with ≈ 50 billion active parameters per token. Built on the Ling 2.0 architecture, Ling-1T is designed to push the limits of efficient reasoning and scalable cognition.
Pre-trained on 20 trillion+ high-quality, reasoning-dense tokens, Ling-1T-base supports up to 128K context length and adopts an evolutionary chain-of-thought (Evo-CoT) process across mid-training and post-training. This curriculum greatly enhances the model’s efficiency and reasoning depth, allowing Ling-1T to achieve state-of-the-art performance on multiple complex reasoning benchmarks—balancing accuracy and efficiency.
We comprehensively evaluated Ling-1T against leading flagship models, including both open-source giants (e.g., DeepSeek-V3.1-Terminus, Kimi-K2-Instruct-0905) and closed-source APIs (GPT-5-main, Gemini-2.5-Pro). Across code generation, software development, competition-level mathematics, professional math, and logical reasoning, Ling-1T consistently demonstrates superior complex reasoning ability and overall advantage.
In the AIME 25 benchmark, Ling-1T extends the Pareto frontier of reasoning accuracy vs. reasoning length, showcasing its strength in “efficient thinking and precise reasoning.”
Ling-1T excels in visual reasoning and front-end code generation tasks, combining deep semantic understanding with precise code synthesis. We introduce a hybrid Syntax–Function–Aesthetics reward mechanism, enabling the model to not only generate correct and functional code but also demonstrate a refined sense of visual aesthetics. On ArtifactsBench, Ling-1T ranks first among open-source models, and the benchmark visualizations in this card were, in fact, generated by Ling-1T itself.
Scaling to the trillion-parameter level has revealed strong emergent reasoning and transfer capabilities. For example, in the BFCL V3 tool-use benchmark, Ling-1T achieves ≈ 70% tool-call accuracy with only light instruction tuning—despite having seen no large-scale trajectory data during training. Ling-1T can:
These capabilities form the foundation for general, collaborative human–AI intelligence, which we aim to advance together with the open-source community through Ling-1T’s release.
The Ling 2.0 architecture was designed from the ground up for trillion-scale efficiency, guided by the Ling Scaling Law (arXiv:2507.17702). This ensures architectural and hyperparameter scalability even under 1e25–1e26 FLOPs of compute.
Key architectural innovations include:
Ling-1T is the largest FP8-trained foundation model known to date. FP8 mixed-precision training yields 15%+ end-to-end speedup, improved memory efficiency, and maintains ≤ 0.1% loss deviation from BF16 across 1T tokens. A fine-grained, heterogeneous 1F1B interleaved pipeline further boosts utilization by 40 %+. System-level optimizations—fused kernels, communication scheduling, recomputation, checkpointing, simulation, and telemetry—ensure stable trillion-scale training.
Pre-training used over 20T high-quality tokens, with > 40% reasoning-dense data in later stages. Mid-training introduced curated chain-of-thought corpora for “reasoning pre-activation”, improving downstream reasoning stability. A custom WSM (Warmup–Stable–Merge) LR scheduler(arXiv:2507.17634) with mid-train checkpoint merging simulates LR decay and boosts generalization.
Built upon mid-training reasoning activation, post-training adopts Evo-CoT (Evolutionary Chain-of-Thought) for progressive reasoning enhancement under controllable cost. This approach continually expands the Pareto frontier of reasoning accuracy vs. efficiency—ideal for reflexive non-thinking models.
For reinforcement learning, we introduce LPO (Linguistics-Unit Policy Optimization) —a novel sentence-level policy optimization method. Unlike GRPO (token-level) or GSPO (sequence-level) algorithms, LPO treats sentences as the natural semantic action units, enabling precise alignment between rewards and reasoning behavior. Empirically, LPO offers superior training stability and generalization across reasoning tasks.
Ling-1T has been extensively evaluated across knowledge, code, math, reasoning, agent, and alignment benchmarks. It currently stands as the best open-source flagship non-thinking model, rivaling closed-source APIs in complex reasoning while maintaining exceptional efficiency and interpretability.
You can download Ling-1T from the following table. If you are located in mainland China, we also provide the model on ModelScope.cn to speed up the download process.
| Model | Context Length | Download |
|---|---|---|
| Ling-1T | 32K -> 128K (YaRN) | 🤗 HuggingFace 🤖 ModelScope |
Note: If you are interested in previous version, please visit the past model collections in Huggingface or ModelScope.
You can experience Ling-1T online at: ZenMux
You can also use Ling-1T through API calls:
from openai import OpenAI
# 1. Initialize the OpenAI client
client = OpenAI(
# 2. Point the base URL to the ZenMux endpoint
base_url="https://zenmux.ai/api/v1",
# 3. Replace with the API Key from your ZenMux user console
api_key="<your ZENMUX_API_KEY>",
)
# 4. Make a request
completion = client.chat.completions.create(
# 5. Specify the model to use in the format "provider/model-name"
model="inclusionai/ling-1t",
messages=[
{
"role": "user",
"content": "What is the meaning of life?"
}
]
)
print(completion.choices[0].message.content)
We will later submit our model to the SGLang official release. Now we can prepare the environment by following these steps:
pip3 install -U sglang sgl-kernel
Both BF16 and FP8 models are supported by SGLang now. It depends on the dtype of the model in ${MODEL_PATH}.
Here is the example to run Ling-1T with multiple GPU nodes, where the master node IP is ${MASTER_IP} and server port is ${PORT}:
# Node 0:
python -m sglang.launch_server --model-path $MODEL_PATH --tp-size 8 --pp-size 4 --dp-size 1 --trust-remote-code --dist-init-addr $MASTER_IP:2345 --port $PORT --nnodes 4 --node-rank 0
# Node 1:
python -m sglang.launch_server --model-path $MODEL_PATH --tp-size 8 --pp-size 4 --dp-size 1 --trust-remote-code --dist-init-addr $MASTER_IP:2345 --port $PORT --nnodes 4 --node-rank 1
# Node 2:
python -m sglang.launch_server --model-path $MODEL_PATH --tp-size 8 --pp-size 4 --dp-size 1 --trust-remote-code --dist-init-addr $MASTER_IP:2345 --port $PORT --nnodes 4 --node-rank 2
# Node 3:
python -m sglang.launch_server --model-path $MODEL_PATH --tp-size 8 --pp-size 4 --dp-size 1 --trust-remote-code --dist-init-addr $MASTER_IP:2345 --port $PORT --nnodes 4 --node-rank 3
# This is only an example. Please adjust arguments according to your actual environment.
curl -s http://${MASTER_IP}:${PORT}/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model": "auto", "messages": [{"role": "user", "content": "What is the capital of France?"}]}'
More usage can be found here
pip install vllm==0.11.0
Here is the example to deploy the model with multiple GPU nodes, where the master node IP is ${MASTER_IP}, server port is ${PORT} and the path of model is ${MODEL_PATH}:
# step 1. start ray on all nodes
# step 2. start vllm server only on node 0:
vllm serve $MODEL_PATH --port $PORT --served-model-name my_model --trust-remote-code --tensor-parallel-size 32 --gpu-memory-utilization 0.85
# This is only an example, please adjust arguments according to your actual environment.
To handle long context in vLLM using YaRN, we need to follow these two steps:
rope_scaling field to the model's config.json file, for example:{
...,
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"type": "yarn"
}
}
--max-model-len to specify the desired maximum context length when starting the vLLM service.For detailed guidance, please refer to the vLLM instructions.
While Ling-1T has made strong progress in efficient reasoning, cross-domain generalization, and training efficiency, several limitations remain:
The future versions of Ling-1T will continue to evolve in architecture, reasoning, and alignment, advancing the series toward more general intelligence.
This code repository is licensed under the MIT License.
Recommended temperature? 0.7
Recommended top_p? 0.95
The public training-content summary identifying Ling-1T is available below. Please refer to the document for its covered model versions, training-content scope, summary version, and update date.
This summary concerns training-content disclosure; it does not replace the model’s technical documentation, usage terms, or license.
If you find our work helpful, feel free to give our paper a citation.
@article{Ling-Team2025,
author = {Ling-Team and 141 others},
title = {{Every Activation Boosted: Scaling General Reasoner to 1 Trillion Open Language Foundation}},
journal = {arXiv preprint arXiv:2510.22115},
eprint = {2510.22115},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
year = {2025},
doi = {10.48550/arXiv.2510.22115},
url = {https://arxiv.org/abs/2510.22115}
}

model-00015-of-00155.safetensors 17 GBmodel-00017-of-00155.safetensors 17 GBmodel-00019-of-00155.safetensors 17 GBmodel-00021-of-00155.safetensors 17 GBmodel-00023-of-00155.safetensors 17 GBmodel-00025-of-00155.safetensors 17 GBmodel-00027-of-00155.safetensors 17 GBmodel-00029-of-00155.safetensors 17 GBmodel-00031-of-00155.safetensors 17 GBmodel-00033-of-00155.safetensors 17 GBmodel-00035-of-00155.safetensors 17 GBmodel-00037-of-00155.safetensors 17 GBmodel-00039-of-00155.safetensors 17 GBmodel-00041-of-00155.safetensors 17 GBmodel-00043-of-00155.safetensors 17 GBmodel-00045-of-00155.safetensors 17 GBmodel-00047-of-00155.safetensors 17 GBmodel-00049-of-00155.safetensors 17 GBmodel-00051-of-00155.safetensors 17 GBmodel-00053-of-00155.safetensors 17 GBmodel-00055-of-00155.safetensors 17 GBmodel-00057-of-00155.safetensors 17 GBmodel-00059-of-00155.safetensors 17 GBmodel-00061-of-00155.safetensors 17 GBmodel-00063-of-00155.safetensors 17 GBmodel-00065-of-00155.safetensors 17 GBmodel-00067-of-00155.safetensors 17 GBmodel-00069-of-00155.safetensors 17 GBmodel-00071-of-00155.safetensors 17 GBmodel-00073-of-00155.safetensors 17 GBmodel-00075-of-00155.safetensors 17 GBmodel-00077-of-00155.safetensors 17 GBmodel-00079-of-00155.safetensors 17 GBmodel-00081-of-00155.safetensors 17 GBmodel-00083-of-00155.safetensors 17 GBmodel-00085-of-00155.safetensors 17 GBmodel-00087-of-00155.safetensors 17 GBmodel-00089-of-00155.safetensors 17 GBmodel-00091-of-00155.safetensors 17 GBmodel-00093-of-00155.safetensors 17 GBmodel-00095-of-00155.safetensors 17 GBmodel-00097-of-00155.safetensors 17 GBmodel-00099-of-00155.safetensors 17 GBmodel-00101-of-00155.safetensors 17 GBmodel-00103-of-00155.safetensors 17 GBmodel-00105-of-00155.safetensors 17 GBmodel-00107-of-00155.safetensors 17 GBmodel-00109-of-00155.safetensors 17 GBmodel-00111-of-00155.safetensors 17 GBmodel-00113-of-00155.safetensors 17 GBmodel-00115-of-00155.safetensors 17 GBmodel-00117-of-00155.safetensors 17 GBmodel-00119-of-00155.safetensors 17 GBmodel-00121-of-00155.safetensors 17 GBmodel-00123-of-00155.safetensors 17 GBmodel-00125-of-00155.safetensors 17 GBmodel-00127-of-00155.safetensors 17 GBmodel-00129-of-00155.safetensors 17 GBmodel-00131-of-00155.safetensors 17 GBmodel-00133-of-00155.safetensors 17 GBmodel-00135-of-00155.safetensors 17 GBmodel-00137-of-00155.safetensors 17 GBmodel-00139-of-00155.safetensors 17 GBmodel-00141-of-00155.safetensors 17 GBmodel-00143-of-00155.safetensors 17 GBmodel-00145-of-00155.safetensors 17 GBmodel-00147-of-00155.safetensors 17 GBmodel-00149-of-00155.safetensors 17 GBmodel-00151-of-00155.safetensors 17 GBmodel-00153-of-00155.safetensors 17 GBmodel-00002-of-00155.safetensors 17 GBmodel-00005-of-00155.safetensors 17 GBmodel-00007-of-00155.safetensors 17 GBmodel-00009-of-00155.safetensors 17 GBmodel-00011-of-00155.safetensors 17 GBmodel-00013-of-00155.safetensors 17 GBmodel-00003-of-00155.safetensors 11 GBmodel-00024-of-00155.safetensors 10 GBmodel-00034-of-00155.safetensors 10 GBmodel-00044-of-00155.safetensors 10 GBmodel-00054-of-00155.safetensors 10 GBmodel-00064-of-00155.safetensors 10 GBmodel-00074-of-00155.safetensors 10 GBmodel-00084-of-00155.safetensors 10 GBmodel-00094-of-00155.safetensors 10 GBmodel-00104-of-00155.safetensors 10 GBmodel-00114-of-00155.safetensors 10 GBmodel-00124-of-00155.safetensors 10 GBmodel-00134-of-00155.safetensors 10 GBmodel-00014-of-00155.safetensors 10 GBmodel-00144-of-00155.safetensors 10 GBmodel-00016-of-00155.safetensors 8.7 GBmodel-00020-of-00155.safetensors 8.7 GBmodel-00026-of-00155.safetensors 8.7 GBmodel-00030-of-00155.safetensors 8.7 GBmodel-00036-of-00155.safetensors 8.7 GBmodel-00040-of-00155.safetensors 8.7 GBmodel-00046-of-00155.safetensors 8.7 GBmodel-00050-of-00155.safetensors 8.7 GBmodel-00056-of-00155.safetensors 8.7 GBmodel-00060-of-00155.safetensors 8.7 GBmodel-00066-of-00155.safetensors 8.7 GBmodel-00070-of-00155.safetensors 8.7 GBmodel-00076-of-00155.safetensors 8.7 GBmodel-00080-of-00155.safetensors 8.7 GBmodel-00086-of-00155.safetensors 8.7 GBmodel-00090-of-00155.safetensors 8.7 GBmodel-00096-of-00155.safetensors 8.7 GBmodel-00100-of-00155.safetensors 8.7 GBmodel-00106-of-00155.safetensors 8.7 GBmodel-00110-of-00155.safetensors 8.7 GBmodel-00116-of-00155.safetensors 8.7 GBmodel-00120-of-00155.safetensors 8.7 GBmodel-00126-of-00155.safetensors 8.7 GBmodel-00130-of-00155.safetensors 8.7 GBmodel-00136-of-00155.safetensors 8.7 GBmodel-00140-of-00155.safetensors 8.7 GBmodel-00006-of-00155.safetensors 8.7 GBmodel-00010-of-00155.safetensors 8.7 GBmodel-00018-of-00155.safetensors 8.6 GBmodel-00022-of-00155.safetensors 8.6 GBmodel-00028-of-00155.safetensors 8.6 GBmodel-00032-of-00155.safetensors 8.6 GBmodel-00038-of-00155.safetensors 8.6 GBmodel-00042-of-00155.safetensors 8.6 GBmodel-00048-of-00155.safetensors 8.6 GBmodel-00052-of-00155.safetensors 8.6 GBmodel-00058-of-00155.safetensors 8.6 GBmodel-00062-of-00155.safetensors 8.6 GBmodel-00068-of-00155.safetensors 8.6 GBmodel-00072-of-00155.safetensors 8.6 GBmodel-00078-of-00155.safetensors 8.6 GBmodel-00082-of-00155.safetensors 8.6 GBmodel-00088-of-00155.safetensors 8.6 GBmodel-00092-of-00155.safetensors 8.6 GBmodel-00098-of-00155.safetensors 8.6 GBmodel-00102-of-00155.safetensors 8.6 GBmodel-00108-of-00155.safetensors 8.6 GBmodel-00112-of-00155.safetensors 8.6 GBmodel-00118-of-00155.safetensors 8.6 GBmodel-00122-of-00155.safetensors 8.6 GBmodel-00128-of-00155.safetensors 8.6 GBmodel-00132-of-00155.safetensors 8.6 GBmodel-00138-of-00155.safetensors 8.6 GBmodel-00142-of-00155.safetensors 8.6 GBmodel-00146-of-00155.safetensors 8.6 GBmodel-00148-of-00155.safetensors 8.6 GBmodel-00150-of-00155.safetensors 8.6 GBmodel-00152-of-00155.safetensors 8.6 GBmodel-00154-of-00155.safetensors 8.6 GBmodel-00008-of-00155.safetensors 8.6 GBmodel-00012-of-00155.safetensors 8.6 GBmodel-00155-of-00155.safetensors 4.6 GBmodel-00004-of-00155.safetensors 3.3 GBmodel-00001-of-00155.safetensors 2.6 GB