default-group API key in the Anthropic Messages or the OpenAI-compatible format, or point Claude Code at LaoZhang API.
- Published: October 18, 2026
- Last verified: November 10, 2026
- Status: active; call them with a
default-group token.claude-fable-5-2was added on November 10, andclaude-sonnet-4-5-20250929was withdrawn on November 4
claude-fable-5-2) is available at $8 input and $40 output per 1M tokens, the same as Anthropic’s list price and 20% below Fable 5.1. Set model to claude-fable-5-2; the endpoints and your token stay the same. For request examples and the request differences, see Claude Fable 5.2 API.
Update, November 4, 2026: Claude Code and other Claude-protocol tools now use a Claude Code-group token (same prices as default); the OpenAI-compatible format still uses default. See Claude Code token group now open.
Update, November 4, 2026: claude-sonnet-4-5-20250929 (Claude Sonnet 4.5) is withdrawn, and requests for it now return 503. Anthropic has deprecated this model and retires it from its API on November 30, 2026. Change model to claude-sonnet-5-5 (input $2, output $10, cheaper than Sonnet 4.5) or claude-sonnet-4-6 (same price as Sonnet 4.5). See the FAQ below.
Key details
Models and prices
Prices from LaoZhang API’s public pricing as of October 18, 2026 (claude-fable-5-2 as of its November 10 listing), in US dollars per 1M tokens:
- Prompt caching is supported. Cache-read prices are in the model catalog, and cache writes cost 1.25× the input price.
- When a
claude-haiku-5-5prompt exceeds 100,000 tokens, counting cache reads and writes, the whole request is billed at $0.5 input and $2.5 output. See tiered pricing in the catalog. - Current prices are in the model catalog and on the console pricing page.
Make a request
Create an API key in Token management with thedefault token group and the Usage first (recommended) or Usage-based billing mode, then export it:
- Anthropic Messages
- OpenAI-compatible
- In the Anthropic Messages format, the answer is in the
content[]blocks whosetypeistext. In the OpenAI-compatible format, it is inchoices[0].message.content. - For streaming, tool use, and prompt caching, see the Claude protocol reference.
- For a complete Claude Code configuration, see Claude Code setup with LaoZhang API.
- Check token usage and the charge for each request in call logs.
Current limits
Check these differences from Anthropic’s own API before you switch:- Each response is capped at 32,768 output tokens. A larger
max_tokensis reduced to the cap. - Through the OpenAI-compatible format (Chat Completions) with no
max_tokens, a response stops at 4,096 tokens, thinking included. Setmax_tokensfor long output, or the reply can be cut off or come back empty. See Why does Claude return an empty reply through Chat Completions? documentcontent blocks, such as PDFs, are not supported. Extract the text and send it instead.- Anthropic server tools such as
web_search,web_fetch, andcode_executionare not supported, and requests that declare them return 400. Claude Code’s WebSearch tool doesn’t work for the same reason. /v1/messages/count_tokens, Message Batches, and the Files API are not available.
Who this affects
- Applies to: projects that call Claude models through LaoZhang API, and anyone using Claude Code or the Anthropic SDKs. Any API key in the
defaultgroup with the Usage first or Usage-based billing mode can call these models. Tokens created earlier in theClaude Codegroup (formerly claude_code) keep working. - Unchanged: no existing model changed. Integrations that call
glm-5.2ordeepseek-v4-proover Anthropic Messages keep working as they are.
FAQ
How is Claude Fable 5.2 different from Fable 5.1?
Fable 5.2 costs less: $8 input, $40 output, and $0.2 for cache reads per 1M tokens, against $10, $50, and $0.25 for Fable 5.1. The endpoints and token are the same, so you only changemodel to claude-fable-5-2. For thinking depth, parameters to leave out, and request examples, see Claude Fable 5.2 API.
Why does claude-haiku-4-5 return a 503?
claude-haiku-4-5-20251001 is not one of the 11 models listed. Use claude-haiku-5-5 instead. Claude Code runs background tasks on that model by default, so set ANTHROPIC_DEFAULT_HAIKU_MODEL as shown in Claude Code setup with LaoZhang API.
Why does claude-sonnet-4-5-20250929 return 404 or 503?
This model was withdrawn on November 4, 2026. Requests for it now return 503 (no available channel); in the days before that they may have returned 404model_not_found. Anthropic has deprecated Claude Sonnet 4.5 and retires it from its API on November 30, 2026, so it will not come back. Change only the model in your request; the endpoint and other parameters stay the same:
claude-sonnet-5-5: Anthropic’s recommended replacement, input $2 and output $10, 1M-token context.claude-sonnet-4-6: the same generation as Sonnet 4.5, at the same price (input $3, output $15).
ANTHROPIC_MODEL or another environment variable still hard-codes claude-sonnet-4-5-20250929.
What should I do about an occasional 529?
A 529 means upstream capacity is busy, which can happen at peak times. Retry after a short wait; the Anthropic SDKs retry 529 automatically by default.Do I need an Anthropic account?
No. Your LaoZhang API key and base URL are all you need.Sources and related links
- Anthropic pricing: official list prices and cache multipliers.
- Claude Fable 5.2 API: pricing, examples, and Fable 5.1 comparison
- Anthropic model deprecations: deprecation and retirement dates for Claude Sonnet 4.5.
- Model catalog: current LaoZhang API prices.
- Console pricing: sign in to see the prices that apply to your account.
- Claude protocol: Anthropic Messages API
- Claude Code setup with LaoZhang API
- Latest announcements