FreeLLMAPI v0.13.1 fixes "Rate limited" errors when you use Claude Code, Claude Desktop and other coding agents through the router.
Fixes
- Claude Code and other agents no longer hit false "Rate limited" errors. Coding agents send large tool definitions with every message, and the router left them out when sizing a request. It could send an agent's request to a model too small for it, such as Groq's 8k-per-minute models. The provider turned it down as too large, and the router reported that back as a rate limit. The router now counts tool definitions and sends each request only to models that can take it. This covers Claude Code, Claude Desktop, OpenAI-compatible agents, the Responses API and the Gemini/Ollama endpoints. (#1362)
- Groq errors no longer read as a used-up daily quota. Every Groq limit error ends with "Upgrade to Dev Tier today", which the router mistook for a daily limit. Groq's per-minute limits now recover within the minute. (#1362)
- A request that is too large no longer benches the model. The model is skipped for that one request only and stays available for everything else. Before, one large agent request could pile up penalties across your models. (#1362)
/v1/messages/count_tokensnow includes tool definitions in its count.
Updating
Download the installer for your platform below, or use ghcr.io/tashfeenahmed/freellmapi:v0.13.1 for Docker. The latest Docker tag also points to this release. Desktop users on v0.13.0 will see this update in the Updates panel.
Existing keys and profiles are retained. This release has no database migrations.
⭐ Like the free router? Go Premium, the live signed catalog, $19/yr, cancel anytime.