[MODELS]

Model catalog

Every model is served through the same OpenAI-compatible endpoint. Models marked demo are available now with limited capacity; signing in adds you to the waitlist for upcoming releases.

DeepSeek V4 Pro

DEMO
1.6T MoE · 49B active1M context

The strongest open MoE of the V4 family. Reasoning is available on demand and is off by default for predictable latency.

"model": "deepseek-ai/deepseek-v4-pro"

DeepSeek V4 Flash

DEMO
284B MoE · 13B active1M context

V4 speed tier with the same optional reasoning. Increase reasoning_effort per request when you need more depth.

"model": "deepseek-ai/deepseek-v4-flash"

GLM 5.2

DEMO
753B MoE · MIT license1M context

The fastest-adopted open release of 2026. Strong coding and agentic performance per dollar.

"model": "z-ai/glm-5.2"

Kimi K2.6

DEMO
1T MoE · 32B active256K context

Moonshot's K-series workhorse, in the top five open models by traffic since launch.

"model": "moonshotai/kimi-k2.6"

Kimi K3

COMING SOON
2.8T MoE · largest open model ever1M context

Open-weights release expected soon. Every account is on the waitlist for access at launch.

Qwen 3.8

COMING SOON
2.4T multimodalTBA

Announced at WAIC Shanghai. On the roadmap pending open release.

Get your API keyread the API docs →