LiteLLM 1.106.1 released
A patch release fixing issues with Anthropic thinking blocks, tool calls, batch access, vector store metadata, token counting, and MCP tool calls.
What changed in 1.106.1
- Anthropic thinking blocks and bindings now preserved correctly on multi-turn requests
- Tool calls from gpt-5.4+ models now appear in the correct response position
- Team keys can now retrieve, cancel and download their own batches without errors
- Vector store metadata updates no longer fail with HTTP 400 errors
- Token counting accuracy improved for PDFs, system prompts, tools and Google requests; MCP bare tool names now work correctly
Check before you update
- Run db_scripts/strip_vector_store_internal_metadata.py on vector stores created before this release
Summarised by AgentGid from BerriAI's official release notes on GitHub, as reported there; read the original before relying on it. Price, status and release dates come from our daily data.
LiteLLM at a glance
| What it is | An AI gateway and Python SDK that calls 100+ model APIs through one OpenAI-style interface, with spend tracking, budgets, rate limits, fallbacks, guardrails and logging. |
|---|---|
| Latest version | 1.104.5, Oct 21, 2026. This page is about an earlier release. |
| Release pace | 36 stable releases in the last 30 days; 1.106.1 came the same day as the previous one |
| Price | Free (OSS). The SDK and proxy are open source and free. An enterprise edition with SSO, audit logs and support is sold separately. Official pricing ↗ (checked Oct 9, 2026) |
| Downloads, last 7 days | 22.9M up 4.3% |
| Gid Score | 72 #2 in Agent tooling for cost, context and memory |
