# Codex Auto Review

The review model behind Codex's "Approve for me": the kernel asks it whether each risky action is safe and only interrupts you when it judges an action dangerous. Officially marked as an internal model; 272K context, with tools, image input, web search and image generation ready to use.

- Model ID (the request body's `model` field): `codex-auto-review`
- Category: Text chat
- Page: https://qd.zsopc.com/en/models/codex-auto-review
- Endpoint: `https://qd.zsopc.com/v1/chat/completions`
- Context window: 272,000 tokens
- Default output limit per request: 128,000 tokens
- Capabilities: Image input · Tool calling · Thinking mode · Web search · JSON mode · Image generation
- Compatible protocols: OpenAI Chat Completions · OpenAI Responses
- Released: 2026-09-09
- Last updated: 2026-11-17

## Pricing

- 70% OFF right now; these are the prices charged:
  - Input: 24.26 credits (≈ ¥0.404 / ≈ $0.06) per 1M tokens
  - Output: 145.58 credits (≈ ¥2.43 / ≈ $0.36) per 1M tokens
  - Cached input: 2.43 credits (≈ ¥0.04 / ≈ $0.006) per 1M tokens
  - Per output image: 80.88 credits (≈ ¥1.35 / ≈ $0.2)
- List price (applies again when the offer ends):
  - Input: 80.88 credits (≈ ¥1.35 / ≈ $0.2) per 1M tokens
  - Output: 485.28 credits (≈ ¥8.09 / ≈ $1.2) per 1M tokens
  - Cached input: 8.09 credits (≈ ¥0.135 / ≈ $0.02) per 1M tokens
  - Per output image: 80.88 credits (≈ ¥1.35 / ≈ $0.2)

Credits are the platform's only unit: 60 credits = ¥1 (machine-readable: `GET https://qd.zsopc.com/api/v1/config/currency`).
Each call first holds the most it could cost, then settles on actual usage and refunds the difference.

## Example call

```bash
curl https://qd.zsopc.com/v1/chat/completions \
  -H "Authorization: Bearer $PLATFORM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"codex-auto-review","messages":[{"role":"user","content":"Hello"}]}'
```

## FAQ

### How do I call Codex Auto Review?

Set the model field of the request body to codex-auto-review, send it to https://qd.zsopc.com/v1/chat/completions, and put your key (it starts with sk-gpushare-) in the Authorization: Bearer header. The official OpenAI, Anthropic and Google Gemini SDKs all work directly; just point base_url at this gateway.

### How much does the Codex Auto Review API cost?

70% OFF right now; the prices charged are below; Input: 24.26 credits (≈ ¥0.404 / ≈ $0.06) per 1M tokens; Output: 145.58 credits (≈ ¥2.43 / ≈ $0.36) per 1M tokens; Cached input: 2.43 credits (≈ ¥0.04 / ≈ $0.006) per 1M tokens; Per output image: 80.88 credits (≈ ¥1.35 / ≈ $0.2); List price (applies again when the offer ends): Input: 80.88 credits (≈ ¥1.35 / ≈ $0.2) per 1M tokens; Output: 485.28 credits (≈ ¥8.09 / ≈ $1.2) per 1M tokens; Cached input: 8.09 credits (≈ ¥0.135 / ≈ $0.02) per 1M tokens; Per output image: 80.88 credits (≈ ¥1.35 / ≈ $0.2). Prices are in credits: 60 credits = ¥1. Each call first holds the most it could cost, then settles on actual usage and refunds the difference.

### How large is Codex Auto Review's context window?

272,000 tokens, with a default output limit of 128,000 tokens per request.

### What can Codex Auto Review do?

Image input, Tool calling, Thinking mode, Web search, JSON mode, Image generation. Compatible protocols: OpenAI Chat Completions, OpenAI Responses.

## Related

- All models and live prices: https://qd.zsopc.com/en/models
- Quickstart: https://qd.zsopc.com/en/docs/quickstart
- FAQ: https://qd.zsopc.com/en/docs/faq
- OpenAPI: https://qd.zsopc.com/openapi.json
