Token Factory

wylon Token Factory

Inference token services for leading open-source SOTA models. Production-ready chat, image, and video models are available through one API key, with request formats compatible with OpenAI, Anthropic, OpenAI Images, and the Volcengine Ark video generation API.

Services

Three generation capabilities, one Token service

Chat, image, and video generation all run on Token Factory, sharing one account, API key, usage view, and billing system for a consistent integration experience.

Chat and agents

Built for complex tasks and agent workflows

Supports intelligent chat, enterprise knowledge bases, tool calling, and long-context workloads.

API compatibilityOpenAI / Anthropic

Image generation

Built for visual content production at scale

Supports image-generation workflows for marketing assets, product content, and creative visuals.

API compatibilityOpenAI Images

Video generation

Built for dynamic content and creative expression

Supports product demos, short-form video, and dynamic creative workflows.

Request formatVolcengine Ark API compatible

Features

Domestic inference infrastructure for production

Model capabilities run on domestic GPUs and the wylon super-node architecture.

Native domestic GPU support

Full-stack support for domestic GPUs including Biren, MetaX, Iluvatar CoreX, and Cambricon, with kernels, engine scheduling, and parallel strategies tuned to each architecture for stronger inference performance on leading models.

New super-node architecture

wylon is among the first providers to deploy inference cloud infrastructure at scale with super-nodes as the minimum compute unit and high-bandwidth interconnects across accelerators. MoE, long-context, and multi-replica workloads benefit from higher communication efficiency, throughput, and tail latency.

System-level cache management

The cache engine spans global request scheduling and automatically reuses system prompts, retrieved context, tool definitions, and multi-turn conversations. Time to first token can improve by up to 10× for long-context agent workloads.

Capabilities

Fast integration with visible usage

Compatible with mainstream APIs

Use one endpoint across multiple model families and switch inference platforms without changing your existing integration protocol.

Supports mainstream agents

Works with widely used agent frameworks and can serve directly as the model backend for coding agents and harnesses.

System-level caching

Cross-request and cross-session context caching reduces latency and repeated token costs.

End-to-end observability

TTFT, TPS, cache hit rate, and detailed token usage are visible in real time.

Keep the API patterns you already use

Each capability follows its corresponding mainstream interface to reduce migration work.

One key, three request types

Use the same account and API key, then replace the service URL and model ID in your existing client.

Chat → OpenAI / Anthropic compatible
Image → OpenAI Images compatible
Video → Volcengine Ark video API request format

# pip install openai
                from openai import OpenAI

                client = OpenAI(
                base_url="https://api.wylon.cn/v1",
                api_key="wl-live-...",
                )

                resp = client.chat.completions.create(
                model="ZhipuAI/GLM-5.3",
                messages=[{"role": "user",
                "content": "Hello, wylon."}],
                stream=True,
                )

                for chunk in resp:
                print(chunk.choices[0].delta.content, end="")
              

# pip install openai
                from openai import OpenAI

                client = OpenAI(
                base_url="https://api.wylon.cn/v1",
                api_key="wl-live-...",
                )

                image = client.images.generate(
                model="Qwen/Qwen-Image-2512",
                prompt="A futuristic cloud computing center",
                size="1664x928",
                )

                print(image.data[0].url)
              

# Volcengine Ark video API-compatible request body
                {
                "model": "lightx2v/LightWan2.2-A14B",
                "content": [
                {
                "type": "text",
                "text": "A product slowly rotates through soft light"
                }
                ],
                "resolution": "720P",
                "duration": 5
                }
                # See the API docs for task submission and status polling

Additional features

Reliable service, high quotas, broad compatibility

Enterprise-grade

Highly available architecture · Reliable service quality

10×

Faster time to first token with high cache hit rates

1M+

Tokens processed per minute at higher tiers

Domestic GPU brands supported

Get started

Start generating with one API key.

FAQ

Which models are supported?

Live models include GLM-5.3, DeepSeek-V4.1-Flash, Qwen-Image-2512, SenseNova-U1.5-8B-MoT, MiniMax H3, and Wan2.2-T2V-A14B. See the model catalog for capabilities and pricing.

Which APIs are compatible?

Chat supports OpenAI and Anthropic compatibility, image supports OpenAI Images, and video uses a Volcengine Ark video API-compatible request format. Refer to the API docs for exact fields and endpoints.

Do you offer enterprise services?

Yes. We offer dedicated plans for enterprise customers. Reach out via Contact us to talk to our solutions team.

How does wylon handle my data?

wylon handles your data in accordance with applicable regulations. We do not use your data to train models or for unrelated commercial purposes. You can request deletion at any time. Details: Privacy Policy.

What if the model I need is not listed?

We continuously expand the curated catalog. Enterprise customers can contact us for model allowlisting, custom onboarding, or dedicated deployment.