---
name: deepnexus-models
description: Delegate sub-tasks to, or get a second opinion from, any DeepNexus model (DeepSeek, GLM, Kimi, Qwen, MiniMax, MiMo and more) through the OpenAI-compatible API. Use for cheap bulk work (summaries, classification, translation), cross-checking an answer with a different model family, or when the user names a specific DeepNexus model.
---

# DeepNexus models

Requires `DEEPNEXUS_API_KEY` in the environment. If the DeepNexus MCP server is connected, prefer its `chat` tool; otherwise call the HTTP API.

## Pick a model

| Need | Model |
|---|---|
| Cheapest fast work: summaries, extraction, translation, classification | `deepseek-v4.1-flash`, `glm-5.3-flash` |
| Hard reasoning, code review, second opinion | `deepseek-v4-pro`, `kimi-k3`, `glm-5.3` |
| Images | `mimo-v2.6-pro` |
| Unsure | `auto` |

List all models and prices: `GET $DEEPNEXUS_BASE_URL/models` (default base `http://localhost:3000/v1`).

## Call

```bash
curl -s "${DEEPNEXUS_BASE_URL:-http://localhost:3000/v1}/chat/completions" \
  -H "Authorization: Bearer $DEEPNEXUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --arg p "$PROMPT" '{model:"deepseek-v4-pro",messages:[{role:"user",content:$p}],max_tokens:2000}')" \
  | jq -r '.choices[0].message.content'
```

## Patterns

- **Second opinion**: send the problem and your proposed answer, ask the other model to find mistakes only. Report agreements and disagreements; do not silently replace your answer.
- **Fan-out**: split large inputs into chunks, process each with a flash model, then merge the results yourself.
- **Budget**: check `GET /balance` before large batches and tell the user the estimated cost (tokens × price per 1M from `/models`).

Never send secrets, credentials or private user data to another model unless the user has agreed.