Google ADK: what’s new in v2.13.0 and our verdict

Our verdict

Update with care: a big, useful release for Gemini users, but it includes a telemetry default change.

Tested. Tested 2026-10-25 (America/Chicago). Installed google-adk 2.13.0 into a fresh Python 3.13 venv (plus litellm for a non-Gemini model). Ran an Agent with a Python function tool on InMemoryRunner against a local Ollama model: the agent called word_count, got {'count': 6} and answered 6, which is correct. Not tested: Gemini models, live voice, MongoDB sessions, the 2.13.0 telemetry default change or the vendor's 8x claim. Test environment: Linux x86_64 box with 8 CPUs and no GPU, no vendor account or API key. Agent tools were pointed at a local open-weight model (Qwen2.5-3B-Instruct, 4-bit, served by Ollama 0.40.11), so we tested the tool, not any vendor's hosted models. Extended 2026-10-25: Google ADK (Python) 2.13.0 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.7 s. Full log: /tests/google-adk.

Full test log with exact commands, versions and raw output: Google ADK (Python) 2.13.0: our test log.

Evidence

Tested (smoke test, 2026-10-25, local Qwen2.5-3B-Instruct, not vendor models). A smoke test is not a benchmark.

What we ran

Installed google-adk 2.13.0 (plus litellm for a non-Gemini model) and ran an Agent with one Python function tool on InMemoryRunner; the agent called the tool and answered 6 (correct). Extended: Google ADK (Python) 2.13.0 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.7 s.

Model used

Qwen2.5-3B-Instruct (4-bit), a small local open-weight model served by Ollama 0.40.11; no vendor-hosted model was used

Environment

Linux x86_64 box, 8 CPUs, no GPU; no vendor account, login or API key

Not tested

Gemini models, live voice, MongoDB sessions, the 2.13.0 telemetry default change, the vendor's 8x claim; Any vendor-hosted model: everything ran on a local 3B model, so tool-call reliability with frontier models is not measured here; Multi-agent flows, memory, long conversations, streaming, tracing and cloud deployment; Anything beyond one tool and one question; three trials is a smoke test, not a benchmark.

Date

Oct 25, 2026 (America/Chicago)

Best for

• Best AI tool for agent building: Apache-2.0 toolkit with evaluation, live audio and new MongoDB services.

What’s new

ADK for Python 2.13.0 promotes 12 experimental features to stable, adds a classifier node for fast workflow routing, live audio with barge-in, and MongoDB session and memory services.

• 2.13.0 (Oct 22): 12 mature experimental features are now stable.
• New TypesafeClassifier and ClassifierNode route workflows with calibrated probabilities in one forward pass instead of an LLM call.
• Live audio interactions with barge-in alongside Gemini speech-to-text and text-to-speech.
• MongoDbSessionService and MongoDbMemoryService (experimental) persist sessions and long-term memory.
• Up to 8x faster execution turns for string-model agents. Breaking: telemetry now defaults to OpenTelemetry GenAI v2.

How to set it up: Install ADK

Install steps come from the vendor's docs. We ran the install and a short real task on 2026-10-25; see the test notes above for what was and was not covered.

1. Install the package

   The README gives pip. If you hit dependency conflicts it also documents a constraints file.

   pip install google-adk

2. Optional extras

   Install the extensions bundle if you need integrations.

   pip install "google-adk[extensions]"

3. Read the 2.13.0 breaking changes

   Telemetry defaults to OpenTelemetry GenAI v2 and removes call_llm. Check the release notes before upgrading production agents.

Pros and cons

Alternatives and comparisons

• OpenAI Agents SDK: Lightweight Python framework for multi-agent workflows, with tools, handoffs, sessions and tracing.
• LangGraph: Low-level framework for building long-running, stateful agents as graphs.
• CrewAI: Python framework for orchestrating role-playing, autonomous AI agents that work as a crew.
• Langflow: MIT-licensed visual builder for AI agents and MCP servers with a very large GitHub following.

See all alternatives to Google ADK

• Semantic Kernel vs Google ADK: which should you use?

Google models we track

• Gemini 4 Argon (released Oct 13, 2026)
• Nano Banana 2.1 (released Oct 6, 2026)
• Gemini 3.8 Flash TTS (released Sep 22, 2026)
• Gemini 3.8 Flash-Lite TTS (released Sep 22, 2026)
• Gemini 3.8 Live (released Sep 15, 2026)
• Gemini 3.8 Live Extended Thinking (released Sep 15, 2026)

All models · Releases by month

FAQ

What we have not verified

• Support for non-Google models
• Real-world performance claims (the 8x figure is the vendor's)

Sources

• ADK 2.13.0 release (fetched 2026-10-25)
• README (fetched 2026-10-25)

Last verified Oct 25, 2026 · Agentic Weekly

Part of the week of Oct 19, 2026 archive.

Related tools

• Langflow: MIT-licensed visual builder for AI agents and MCP servers with a very large GitHub following.
• Botpress: Customer-support AI agent platform with a free plan and conversation-based paid plans.
• Gumloop: Agent builder for teams; plans start at $37 per month with 20k included credits.