Models
The models the gateway serves, with their limits.
hyde/glm-5.3-flash
Fast everyday model
- Context window: 1M tokens (1,048,576, prompt + output) · Max output: 128K tokens
- Entire codebases in one prompt
- Best for: everyday chat, edits, quick answers
hyde/kimi-k3
Vision · reasoning · long-context model
- Context window: 1M tokens (1,048,576, prompt + output) · Max output: up to the full 1M window
- Reads images as well as text; reasons before it answers (low to max effort)
- Best for: reading, image-heavy work, long documents
hyde/kimi-k3.1-pro
Vision · always-on reasoning · long-horizon model
- Context window: 1M tokens (1,048,576, prompt + output) · Max output: 1,048,576 tokens
- Reads images as well as text; always reasons before it answers (low, high, or max effort)
- Images: send as base64-encoded data, not a public URL
- Best for: long-running coding and agent tasks
hyde/qwen3-embedding-8b
Text embeddings (/v1/embeddings)
- Max input: 8,192 tokens per text · Output: 4,096-dimension vectors
- Multilingual; for retrieval, prefix queries with an instruction as the model card describes
- Best for: search, retrieval (RAG), clustering
hyde/nemotron-3.5-embed-8b
Text embeddings (/v1/embeddings)
- Max input: 8,192 tokens per text · Output: 4,096-dimension vectors
- For retrieval, prefix queries with "query: " and documents with "passage: "
- Best for: search, retrieval (RAG), clustering
Tip
The gateway sets no limits of its own. The context window and output ceiling
come from the upstream provider, so set max_tokens as high as the model
allows. Leave it out and the model default applies. A negative value such
as -1, or a value sent as a string, is treated as unset.
Was this page helpful?