03 / MODELS
Models
Every model Nebius Token Factory serves is available in NConnect: GLM 5.3, Kimi K3, DeepSeek V4, Qwen 3.5, MiniMax M3 and more. Compare context windows and vision support.
The model list is fetched live from Nebius at startup, so every model they serve is available and each model's vision support comes from the API's own modality field, never a hand-maintained list. Results are cached locally and fall back to a bundled snapshot when offline.
| Model | Id | Best for | Context | Vision |
|---|---|---|---|---|
 GLM 5.3 FlashDefault | zai-org/GLM-5.3-Flash | Fast, very low cost, agentic | 1M | No |
| GLM 5.3 | zai-org/GLM-5.3 | Coding, reasoning, tool use | 1,024K | No |
| DeepSeek V4 Pro 0813 | deepseek-ai/DeepSeek-V4-Pro-0813 | Reasoning and agentic coding | 979K | No |
 Kimi K3 | moonshotai/Kimi-K3 | Frontier coding + agentic | 1M | No |
| Kimi K2.6 | moonshotai/Kimi-K2.6 | Vision flagship | 262K | Yes |
| Kimi K2.7 Code | moonshotai/Kimi-K2.7-Code | Coding | 262K | No |
 MiniMax M3 | MiniMaxAI/MiniMax-M3 | Fast, cheap | 196K | No |
 Qwen 3.5 397B | Qwen/Qwen3.5-397B-A17B | General / coding flagship | 262K | No |
| DeepSeek V4 Flash | deepseek-ai/DeepSeek-V4-Flash | Fast DeepSeek V4 | 1M | No |
| DeepSeek V4 Pro | deepseek-ai/DeepSeek-V4-Pro | Long-context reasoning | 1M | No |
| Qwen2.5-VL 72B | Qwen/Qwen2.5-VL-72B-Instruct | Vision fallback | 32K | Yes |
Selecting a model
GLM 5.3 Flash is the default. Pass --model before the harness name to pick another one, or switch inside the agent where it supports that.
nconnect --model zai-org/GLM-5.3 codex
nconnect --model deepseek-ai/DeepSeek-V4-Pro-0813 codex
nconnect --model moonshotai/Kimi-K3 hermesProxied harnesses remember the model you last switched to inside a session and use it as the default on the next launch.