Anthropic Claude API Updates, Gemini Live Tools
TL;DR
Today Anthropic's Claude Platform release notes add a between_tools option to the Models API, extend the Compliance API to cover saved custom instructions, and open a new Claude in Amazon Bedrock endpoint. Google also published an Agent Development Kit guide showing how streaming tools and response_scheduling keep Gemini 3.8 Live voice agents talking while slow backend calls run, cutting time to first words from 14.4 seconds to 1.2 seconds in Google's benchmark.
Key takeaways
- Anthropic's Models API now reports a between_tools thinking type that turns off up-front thinking for supported Claude models.
- The Compliance API can now surface the Instructions for Claude saved in claude.ai and the Global instructions saved in Cowork, in beta for Claude Enterprise organizations.
- A new Claude in Amazon Bedrock endpoint at /anthropic/v1/messages uses the same request shape as the first-party Claude API.
- Google's ADK guide shows how streaming tools and response_scheduling keep Gemini 3.8 Live voice agents talking during slow tool calls, cutting time to first words from 14.4 seconds to 1.2 seconds in its benchmark.
- Saved prompts, variables, and evals are not supported in the updated Workbench in the Anthropic Console.
Anthropic: Claude Platform Release Notes
Anthropic's Claude Platform release notes add a between_tools option to the Models API that turns off up-front thinking, and the GET /v1/models and GET /v1/models/{model_id} endpoints now report this capability for each model. The Compliance API now returns the text organization members saved as Instructions for Claude in claude.ai and as Global instructions in Cowork, in beta for Claude Enterprise organizations. A new Claude in Amazon Bedrock endpoint at /anthropic/v1/messages uses the same request shape as the first-party Claude API.
- between_tools thinking type added to the Models API, turns off up-front thinking
- GET /v1/models and GET /v1/models/{model_id} report the new thinking capability
- Compliance API returns saved Instructions for Claude (claude.ai) and Global instructions (Cowork), in beta for Claude Enterprise
- New Claude in Amazon Bedrock endpoint at /anthropic/v1/messages uses the same request shape as the first-party Claude API
Why it matters: Enterprises can now audit exactly what employees configured as Claude's custom instructions, which affects how consistently Claude surfaces information about a brand across an organization's deployments.
Source: Anthropic: Claude Platform release notes - Claude Platform Docs: new entries (2026-10-20)
Google: Gemini 3.8 Live Streaming Tools
Google published an Agent Development Kit (ADK) guide showing how to use streaming tools, non-blocking FunctionTool calls, and response_scheduling so voice agents on Gemini 3.8 Live stay conversational during slow backend calls. In Google's benchmark sales-report workflow on gemini-3.8-live, switching from a blocking tool to a streaming tool cut the time to the agent's first words from 14.4 seconds to 1.2 seconds. The guide is part of the open-source ADK and the Gemini Live API Toolkit.
- Streaming tools, non-blocking FunctionTool calls, and response_scheduling now documented for Gemini 3.8 Live
- Benchmark: time to agent's first words cut from 14.4 seconds to 1.2 seconds
- Built with the open-source Agent Development Kit (ADK)
- Part of the Gemini Live API Toolkit
Why it matters: Brands running voice assistants or support bots on Gemini Live can keep the conversation alive during slow lookups, which matters because customers hang up on dead air before the assistant ever gets to mention or recommend a brand.
Source: Google: Step by Step guide: Build a Voice Agent That Never Goes Silent using ADK (2026-10-19)
What this means for your brand's AI visibility
We watch these platform-level changes because they shape how reliably an assistant can answer a question about a brand in real time. Anthropic's between_tools option and the Bedrock endpoint are mostly developer-facing, but the Compliance API change matters more directly: when an enterprise can see exactly what employees saved as custom instructions for Claude, it becomes easier to audit whether those instructions are steering Claude's answers about a company in a consistent direction.
On the Google side, a voice agent that goes silent for over ten seconds loses the customer before it ever gets to recommend or mention a brand. The fix in this ADK guide, streaming tools that narrate progress instead of freezing, is a reminder that AI visibility is not only about what an assistant says, it is about whether the conversation survives long enough for the assistant to say it. If you want to track how Claude mentions your brand, these API-level changes are worth watching as Claude Enterprise rolls out tighter controls over custom instructions. As Gemini 3.8 Live voice agents become a bigger surface for customer conversations, it helps to track how Gemini mentions your brand across both text and voice answers. These platform updates are a good reminder to revisit your AEO and GEO strategy as voice agents and enterprise-governed assistants take on more of the conversation.
Frequently asked questions
What did Anthropic change in its Claude Platform release notes?
Anthropic added a between_tools option to the Models API that turns off up-front thinking, extended the Compliance API to return saved Instructions for Claude and Cowork Global instructions for Claude Enterprise organizations in beta, and opened a new Claude in Amazon Bedrock endpoint at /anthropic/v1/messages that uses the same request shape as the first-party Claude API.
What is the between_tools thinking type in Claude's Models API?
It's a new value that GET /v1/models and GET /v1/models/{model_id} can report, showing whether a model accepts thinking of type between_tools, which turns off up-front thinking.
What did Google's ADK guide show for Gemini 3.8 Live?
It showed how to use streaming tools, non-blocking FunctionTool calls, and response_scheduling so a Gemini 3.8 Live voice agent keeps talking during slow backend calls; in Google's benchmark this cut the time to the agent's first words from 14.4 seconds to 1.2 seconds.
Are saved prompts, variables, and evals supported in Anthropic's updated Workbench?
No, the release notes state that saved prompts, variables, and evals are not supported in the updated Workbench in the Console.
Related on Sophyx
- track how Claude mentions your brand
- track how Gemini mentions your brand
- AEO and GEO
- Yesterday's update: Anthropic's Benefit Fund and OpenAI's ChatGPT Plugin Growth
Sources are linked above. Written by Reza Kashi from each company's own announcement; we update this brief every weekday morning.
Comments
Loading comments…