Vapi Release Notes

What's New: Week of October 12, 2026

Traffic splitting: You can now split assistant traffic across published versions by percentage when you publish, so you can canary a new version and roll back instantly, with a live traffic view in the assistant header.

Version-aware call logs: You can now filter your call logs by assistant version in each assistant's Logs tab.

GPT-Live keypad input and Google/Slack tools (Beta): GPT-Live calls can now accept incoming keypad (DTMF) input on phone calls, and Google Calendar, Google Sheets, and Slack tools now run on the GPT-Live reasoner.

Squad builder refresh: Selecting a squad assistant now opens the assistant's settings in a side panel on the right. Options and settings for the squad are now in the Talk menu and the More actions (⋮) menu, and you can add an assistant from the left sidebar.

New Models

DeepSeek Flash: DeepSeek's deepseek-flash model is now available for assistants when you connect your DeepSeek account.

Fixes and Improvements

Call log playback: In call detail you can now click a message to play it, with new waveform view options.

Krisp denoiser: Fixed Krisp failing at a 24 kHz sample rate, fixed denoised audio sometimes sounding stretched on websocket calls, and denoiser failures now appear in the call log.

Simulations: Fixed an issue where simulation runs could get stuck reconnecting while catching up on live progress.

What's New: Week of October 5, 2026

• OpenAI GPT-6 Luna: OpenAI's GPT-6 Luna (gpt-6-luna) is now available for assistants in the API and dashboard, with model-specific pricing, reasoning support, and fast service tier.
• ElevenLabs v4 Turbo: The eleven_v4_turbo voice model is now available in the dashboard voice picker and the API.
• Deepgram nova-3-pharma: Deepgram's nova-3-pharma transcription model is now available for assistants.
• Soniox low-confidence filter: You can now set a confidence threshold on the Soniox transcriber to discard low-confidence transcripts, from the transcriber Advanced settings and via the API.
• Simulation latency expectations: Simulations can now set latency limits for model, voice, and turn timing, fail runs that exceed them, and show measured latency results.

Fixes and Improvements

• Visual prompt editor: The visual prompt editor no longer alters nested or multi-line Liquid tags or section headings when switching modes or saving.
• Chat handoff variables: Chat squad handoffs now extract variables from the chat conversation instead of overwriting valid handoff values.

What's New: Week of September 28, 2026

Composer Per-Conversation Approval Modes: Composer now lets you choose when each conversation pauses for approval: ask before all changes, allow creates but ask before updates, deletes, runs, or publishing, bypass approvals entirely, or set custom permissions by resource and action.

Eval Run History: You can now browse your complete Eval run history with search and pagination, and share a specific run result by URL.

Fixes and Improvements:

Call logs: The logs table now shows clearer, deterministic error messages for 4xx responses instead of retrying unnecessarily, and the date picker now lets you select the full retention window.

Voice fallback: xAI terminal voice failures raised before a call starts are now delivered, the fallback voice no longer greets out of turn, and replacing a voice no longer consumes a second fallback.

What's New: Week of August 31, 2026

Campaigns

You can now create and run outbound calling campaigns from the dashboard. Upload your contacts, pick an assistant, and track results as calls complete.

Prompt Editor

The assistant editor now includes a prompt editor where you can generate a prompt from a short description, insert variables, and search within your prompt.

Soniox Endpoint Controls

You can now tune the Soniox transcriber's endpoint sensitivity and endpoint latency adjustment level for finer control over turn taking.

Named Transfer Destinations

You can now give each transfer destination a friendly name that shows in your call logs instead of a raw phone number.

Auto-Start Video Recording

For web calls with video recording enabled, the recording now starts automatically when the caller joins, so your video recordings are captured reliably. This requires upgrading to the latest web SDK version.

What's New: Week of August 17, 2026

Assistant and Tool Versioning: Edits no longer go live immediately. Work in a draft, publish when you're ready, and every publish creates a named version snapshot, so you can go back to your history to see what changed or roll back to a previous version as needed. You can also pin a specific tool version to an assistant.

Call Artifact Upload Webhook: A new call.artifact.upload server message fires as each recording, packet capture, and log finishes uploading, and it includes the result for each, so you can process artifacts as they arrive instead of waiting for the end of call report.

Azure Region Pinning for GPT-5 Models: You can now pin a GPT-5 model to a specific Azure region by adding the region to the model name after a colon (for example gpt-5:eastus2), which keeps requests in that region for lower latency and data residency. It uses Vapi's platform Azure credential, so no Bring Your Own Key (BYOK) is needed.

What's New: Week of August 10, 2026

• Simulations: You can now build simulation suites and run them against your assistants or squads to test and evaluate their behavior, with detailed results for each run.
• Unified Provider and Model Picker: The assistant editor now combines provider and model into a single picker for your LLM, voice, and transcriber.
• Chat and Session Log Export: You can now export selected rows from chat and session logs and use a bulk-actions bar, matching the call logs experience.
• SIP Transfer Improvements: Blind call transfers now support a configurable fallbackPlan, SIP verb, and dial timeout, and cold-transfer outcomes now appear in call logs.
• Composer Improvements: Composer is now more capable, reliable, and easier to use when building voice agents.
• Upload files for it to reference, read current Vapi documentation directly, and follow new Vapi-built skills for building assistants, choosing models, managing tools, configuring phone numbers, and debugging calls.
• Longer, multi-step tasks are far more reliable, with mid-task error recovery and progress that survives page refreshes and thread switches.
• Connect Google Calendar from the conversation, clickable resource mentions with copyable IDs, and cleaner, collapsible activity timelines.

What's New: Week of July 27, 2026

New Models

Anthropic Claude Sonnet 5: Anthropic's Claude Sonnet 5 is now available as a model for assistants.

OpenAI GPT-5.5 "Instant": OpenAI's GPT-5.5 (gpt-5.5 and gpt-5.5-chat-latest) is now available as a model for assistants.

OpenAI GPT-5.6 Models: OpenAI's GPT-5.6 models (sol, terra, and luna) are now available for assistants.

Inworld TTS-2: Inworld's TTS-2 voice is now available for assistants.

SIP Call Transfers

SIP Warm Transfer Audio: Warm call transfers on SIP calls now play hold audio to the operator and a completion sound (transferCompleteAudioUrl) when the transfer connects.

What's New: Week of July 20, 2026

Model Intelligence

You can now set your assistant's transcriber, llm, and voice models in one click with a Model Preset (choose between Balanced, High Intelligence, Ultra Fast, or Cost Saver), and see the latency, cost, and quality metrics for your chosen models so you can compare options and optimize with data.

Recording Download URLs in the End of Call Report

The end of call report now includes short-lived presigned download URLs for your call recordings and logs, so you can download them directly.

End of Call Reports with Zero Data Retention

End of call reports are now reliably delivered for transient assistants running under Zero Data Retention.

What's New: Week of July 13, 2026

Playback Speed for Recordings: Call recording players now include a playback-speed control.

Download Every Recording Type: You can now download every recording a call produced: mono, stereo, separate assistant and customer tracks, video, and packet capture.

VAD Transitions in Call Logs: The call log now surfaces voice-activity-detection transitions with a per-phase latency breakdown.

HIPAA Compliance: xAI is now HIPAA-compliant across its model, voice, and transcriber.

What's New: Week of June 29, 2026

OpenAI Realtime v2: OpenAI's latest Realtime v2 model is now available for assistants.

AI-Generated Tool Failure and Completion Messages: You can now set role: 'system' on request-failed tool messages, and the dashboard has a new UI for configuring AI-generated messages when tools fail or complete. This gives assistants more natural responses when tools hit errors.

Call Logs Improvements:

• Significant latency improvements across both the dashboard and the /calls API endpoints.
• The call detail flyout now shows which assistant or squad handled the call, includes a link to the assistant or squad, and indicates which phone number was used.
• In squad or handoff calls, transcript messages now show which assistant said what, making it easier to trace conversation flow.

MCP Child Tools in Dashboard: When you connect an MCP server, the dashboard tool form now lists all child tools it discovers, so you can see exactly what capabilities your MCP server exposes.