> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.vapi.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.vapi.ai/_mcp/server.

# What's New: Week of October 26, 2026

## New Models and Transcribers

1. **gpt-transcribe (OpenAI)**: OpenAI's [`gpt-transcribe`](/providers/transcriber/openai) transcription model is now available, with support for keywords and a prompt to guide transcription.
2. **AssemblyAI on HIPAA**: [AssemblyAI](/providers/transcriber/assembly-ai) can now be used as a transcriber on HIPAA-enabled assistants.

## Features

1. **Traffic tab**: Each assistant now has a [Traffic tab](/assistants/versioning/traffic-tab) that shows how calls split across published versions, with a call-flow view, a version comparison table, and metrics you can filter by date range and version.
2. **Regional OpenAI models**: You can now select [regional OpenAI models](/providers/model/openai#regional-models) when configuring an assistant, squad, eval, or structured output.

## Fixes and Improvements

1. **GPT-Live compatibility**: [GPT-Live](/gpt-live/configuration) assistants don't support squads or recording consent plans. Previously, a call that used a GPT-Live assistant in a squad, or with a recording consent plan, was created and then ended as soon as it started. These calls are now rejected when they're created, with an error that says what to change.
2. **ElevenLabs voice errors**: When an ElevenLabs voice fails, the call now ends with a specific [reason](/calls/call-ended-reason), such as an invalid API key or an exceeded quota, instead of a generic voice error.