Claude dual-route routing restoration and model catalog synchronization¶

An urgent correction restores reliable handling for Claude models across multiple API protocols, brings back access to claude-fable-5, and ensures strict model catalog synchronization between the API and Telegram bot.

Fixed issues¶

Claude multi-protocol routing¶

Fixed a routing lookup error when accessing Claude models via the OpenAI-compatible POST /v1/chat/completions endpoint.
Decoupled request dispatch between Anthropic Messages (POST /v1/messages) and Chat Completions, restoring reliable connectivity for Claude Code, Codex, and generic OpenAI-compatible client tools.
Eliminated 500 Internal Server Error responses caused by route conflicts on OpenAI-formatted Claude requests.

Availability of claude-fable-5¶

Restored active routing and accurate pricing rules for claude-fable-5.
The model now correctly appears in GET /v1/models and in the Models section of the Telegram bot for authorized API keys.

Catalog visibility and permissions synchronization¶

Aligned catalog filtering logic: GET /v1/models and the Telegram bot now use the same canonical algorithm to verify key permissions.
Inactive models and models not permitted for a given key no longer appear in available model listings.

Improvements¶

Pricing rule evaluation performance¶

Deferred compilation of pricing rules during request dispatch (lazy matching), reducing request overhead and API latency.

No changes to API keys, base URLs, or client libraries are required.
To check active model availability for your key, call GET /v1/models or view the Models screen in the Telegram bot.