Claude Code Discount group: Claude at half price for API calls and Claude Code

The new Claude Code Discount token group bills Claude models at half the catalog price. Use it for direct API calls over the Claude protocol (/v1/messages) or with Claude Code and other Claude-protocol tools, with streaming and tool use; only the key changes. The group runs on a Claude Max subscription route, so teams and production should stay on the Claude Code group.

Published: November 13, 2026
Last checked: November 13, 2026
Status: Active

Key facts

Item	Details
Token group	`Claude Code Discount` in the console (errors show its key, `claude_code_saver`), chosen when you create or edit a token in Token management
Pricing	Half of the model catalog price, for input, output, and cache reads and writes alike
Route	Claude Max subscription route, not the official API
Protocol	Claude protocol (Anthropic Messages), `POST https://api.laozhang.ai/v1/messages`
Billing mode	Usage first (recommended) or Usage-based
Availability	Open to all users

The group serves 11 models, all Claude:

Claude 5.5: claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5
Claude Fable: claude-fable-5-2, claude-fable-5-1, claude-fable-5
Claude 5: claude-opus-5, claude-sonnet-5
Claude 4.x: claude-opus-4-7, claude-opus-4-6, claude-sonnet-4-6

Discount or Claude Code: which group to pick

Item	`Claude Code Discount`	`Claude Code`
Price	Half the catalog price	Full catalog price, same as `default`
Route	Claude Max subscription route	Official API
Models	11 Claude models	Claude, DeepSeek, GLM, Kimi
Best for	Personal projects, development and testing, everyday Claude Code use, tight budgets	Teams, production, behavior that matches the official API

Need DeepSeek, GLM, or Kimi? Use Claude Code.
Calling other models in the OpenAI-compatible format (Chat Completions, Responses)? Use default.

What behaves differently

Replies and billing on the Discount group differ from the official API in a few ways. Switch only if these are fine for you:

The model may introduce itself as Claude Code. Answers lean toward a coding assistant’s style.
Models may think even with thinking off. Opus, Fable, and similar models can think first, and thinking tokens count as output tokens, so a short reply may bill more output tokens than the visible text.
Token counts may differ slightly from Anthropic’s own counts. The usage logs show the actual charge for each request.
Teams and production belong on Claude Code. If you need behavior identical to the official API or have data-sourcing requirements, use the Claude Code group, which runs on the official API.

How to switch

Create a Discount token

In Token management, create a token, choose the Claude Code Discount group and the Usage first billing mode, then copy the key. You can also edit an existing token and change its group.

Direct API calls: swap the key

Keep the endpoint POST https://api.laozhang.ai/v1/messages, the headers, and the request body; only the key changes. With the Anthropic SDK, change api_key and keep base_url as https://api.laozhang.ai. See Claude protocol for full examples.

Claude Code: put the key into settings

Replace ANTHROPIC_AUTH_TOKEN in ~/.claude/settings.json with the new key and keep ANTHROPIC_BASE_URL as https://api.laozhang.ai.Every model variable must name one of the Claude models above, including ANTHROPIC_DEFAULT_HAIKU_MODEL. If any of them still says glm-5.2 or deepseek-v4-flash, change it. See Use Claude models for the settings.

Send a test request

With Claude Code, run claude -p "Reply with exactly one word: connected"; it should print connected. For direct API calls, any request that returns 200 confirms the key. The usage logs show the group and the half-price charge for the request.

Who is affected

Personal projects, development and testing, and Claude Code users who only need Claude models: move to the Claude Code Discount group, for direct API calls or Claude Code, and halve the cost.
Teams, production workloads, and anyone using DeepSeek, GLM, or Kimi: stay on the Claude Code group. Its prices and route don’t change.
default and other groups: nothing changes.

FAQ

How is the Claude Code Discount group billed?

Claude models cost half of their price in the model catalog, and the same rate applies to input, output, cache reads, and cache writes. Billing modes work as in the Claude Code group: Usage first or Usage-based. Check each request’s actual charge in the usage logs.

Why does a Discount token return 503 for DeepSeek or GPT?

The group holds only the 11 Claude models: no DeepSeek, GLM, or Kimi, and no GPT, Gemini, image, or video models. Requests for them return Current group claude_code_saver has no available channels for model …. Use a Claude Code token for DeepSeek, GLM, and Kimi, and a default token for everything else.

Why are output tokens higher than the reply text?

On the Discount group, Opus, Fable, and similar models may think before answering even when thinking is off, and those thinking tokens are billed as output. A short reply can therefore show many more output tokens than its text suggests. If you need output that behaves like the official API, use the Claude Code group.

Can I use a Discount token with Chat Completions?

Plain chat works, but as with the Claude Code group we don’t recommend it. For tool calls, use the Claude protocol at /v1/messages. Send OpenAI-compatible requests with a default token.

Can teams or production services use the Discount group?

We don’t recommend it. The group runs on a Claude Max subscription route, and its replies and billing differ from the official API as listed above. Keep teams, production, and anything that must behave exactly like the official API or has data-sourcing requirements on the Claude Code group, which uses the official API.