Famulor Update May 2026: GPT Realtime 2 & Sonic 3.5

Famulor Product Update May 2026: GPT Realtime 2, Sonic 3.5, GPT-5.5, and Self-Service Telephony

The May 2026 release ships the biggest voice stack upgrade of the year: OpenAI GPT Realtime 2 in speech-to-speech and Dualplex™ mode, Cartesia Sonic 3.5 as the new voice engine, GPT-5.5 in pipeline mode, self-service imports for Twilio, Telnyx, Zadarma, and DIDLogic, post-call re-transcription, a new split-view detail panel, API endpoints for AI pause per conversation, and auto top-up for chat credits. This post walks through every change, when it makes sense, and how to enable it inside your assistant.

Short version: if you run production voice agents on Famulor, your stack benefits immediately from higher instruction-following quality, more natural-sounding speech, lower latency, and easier carrier onboarding — with no configuration overhead. Several upgrades land automatically; others can be toggled per assistant.

Highlights at a glance

🧠 GPT Realtime 2 — Better prompt following in speech-to-speech and Dualplex™

Our speech-to-speech and Dualplex™ mode now runs on OpenAI GPT Realtime 2. The biggest win: significantly more reliable prompt following. The assistant stays cleanly on script, respects instructions more precisely, and holds context stably through multi-step conversations.

What that means in practice — three typical scenarios from the Famulor customer base:

• Scripted lead qualification — sales teams define a fixed conversation path (hello → need → budget → meeting). GPT Realtime 2 strays from that path less often, asks the right slots in the right order, and keeps context through interruptions.
• Compliance-critical workflows — in regulated industries certain disclaimers must be read verbatim (privacy notice, EU AI Act Article 50(3) transparency). The new pipeline swallows those phrasings far less often.
• Multi-step triage — reception flows with three or four nested decisions (e.g., "first ask for patient number, then symptom, then triage") run more stably without loop breaks.

Combined with our proprietary Dualplex™ technology, this delivers the best voice quality, lowest latency, and most reliable instruction-following performance on any AI calling platform. If you already use Dualplex™, you are automatically upgraded — no configuration needed. If you still run a classic pipeline mode, you can switch in the no-code builder.

🎙 Cartesia Sonic 3.5 — Voice engine upgrade, automatically on

The Cartesia voice engine has been updated to Sonic 3.5. Voices sound noticeably more natural, stay stable through long calls, and pronounce difficult words and names more consistently. The upgrade applies automatically to all assistants on Cartesia voices — no configuration changes required.

Three use cases where Sonic 3.5 matters most:

1. Outbound campaigns with long scripts — at 60-seconds-plus pitch lengths, earlier Sonic versions sometimes showed "voice drift." Sonic 3.5 holds the timbre stable across the full duration.
2. Proper nouns and vertical jargon — pharma, legal, real estate each have their own vocabulary. The new engine pronounces those terms more consistently.
3. Multilingual campaigns — in mixed EN/DE calls (e.g., tech support for SaaS), language switches sound cleaner.

You don't have to do anything. If you want to A/B test, pull an old call recording, swap the voice on the assistant, and compare. You'll find voice selection in the usual place — assistant configuration under synthesizer.

⚡️ GPT-5.5 — New language model in pipeline mode

GPT-5.5 is now available as a language model in pipeline mode. A strong all-rounder with fast responses, solid reasoning, and very good prompt following — ideal when you want to combine speed and intelligence without flagship cost. Available immediately in your assistant's pipeline configuration.

When GPT-5.5 is the right choice:

📞 Bring Your Own Telephony — Self-service carrier imports

You can now connect your own carrier accounts directly through a guided import wizard — no support ticket and no manual SIP setup. This was historically the most time-consuming step in onboarding and is now a self-service flow.

• Twilio SIP Trunks — full self-service import with step-by-step onboarding panel. Pricing transparently pre-calculable in the Twilio calculator.
• Telnyx — connection through the same wizard flow as Twilio. Pricing check via Telnyx calculator.
• Zadarma & DIDLogic — two additional carriers in the picker for more region and pricing flexibility, especially attractive for Eastern European and Asian number pools.

Concrete use case: anyone with an existing Twilio account and purchased numbers can port them into Famulor in under 5 minutes — without buying new numbers and without switching carriers. This drastically reduces onboarding friction for enterprise customers locked into long-term telephony contracts.

🎤 Post-call re-transcription for Dualplex™ and Multimodal

Completed calls can now be re-transcribed from the original recording at any time — ideal when you want to re-process with the strongest STT model. Available for Dualplex™ and Multimodal assistants, including a transparent billing view directly in the assistant view.

Three common reasons to re-transcribe a call:

1. QA and coaching — if an auto-generated transcript contains errors, the call can be re-processed with the stronger STT — e.g., for training data extraction.
2. Compliance audit — for GDPR requests or legal disputes, the cleanest possible version of a conversation is often required. Re-transcription delivers exactly that.
3. Feature extraction — if new extracted variables are defined later, re-transcription helps pull them from older conversations.

📂 Split-view detail panels — Calls and Conversations

List views for Calls and Conversations now offer a side-by-side detail panel: select a record on the left, see all details on the right — no more constant back-and-forth. At high call volumes this is a massive workflow accelerator.

• Performance indices — significantly faster navigation between records, even on very large lists.
• Extracted variables section — extracted variables are now shown in a dedicated section of conversation details. Historically buried deep in the UI, now first-class.

If you run daily audits or QA reviews on Famulor, you gain measurable time here. The view is part of all plan tiers — no separate add-on.

💬 Disable and enable AI per conversation

Two new API endpoints allow you to pause and resume AI processing programmatically per individual conversation — perfect for human handoff workflows in WhatsApp and chat. With AI disabled, your team takes over manually; when needed, you re-enable AI for the next handoff.

This is a central feature for customer support teams running hybrid workflows — roughly 80% of simple cases handled by the bot, escalations and VIP cases by humans. Example workflow:

1. Incoming WhatsApp chat — bot handles FAQ.
2. Customer asks for a human agent — webhook triggers POST /conversation/{id}/ai-disable.
3. Conversation lands in the lead Kanban for the human team; the conversation continues entirely manually.
4. Once the issue is resolved, the team calls POST /conversation/{id}/ai-enable — bot takes over routing, follow-up, or closing.

Full endpoint docs are available in the integrations area and the developer API.

💸 Auto top-up for chat credits

Chat credits can now be topped up automatically when the balance runs low. Email notifications transparently track every top-up. This keeps chat experiences uninterrupted, even overnight.

Most useful for teams with:

• High WhatsApp volume — if a campaign suddenly ramps up at night, the bot keeps running without a forced pause.
• Seasonal spikes — Black Friday, Christmas, tax season — no more manual top-ups required.
• Multi-location setups — branches no longer need to handle their own top-ups.

Enable it in the billing area of your account. The threshold is freely configurable — no preset levels.

🔄 Telephony and call-flow improvements

• Initial message on AI-to-AI transfer — the target assistant speaks its initial message immediately for a seamless handoff between two Famulor assistants. Important for multi-bot architectures where a front-desk bot hands off to a specialist bot.
• Transfer caller details visible — transfer target and client phone number are now shown in transfer calls in the call view. This makes debugging and QA of complex routing setups much easier.
• Single-character tool parameter names — tool parameters can now consist of a single character (handy for compact JSON schemas in token-sensitive setups).

🛠 Additional improvements and fixes

• WhatsApp pre-verified number deletion — deletion of pre-verified WhatsApp numbers now works reliably.
• Phone number deletion — deletion jobs for phone numbers run more stably, including correct logging on error.

What you should do today

1. Audit your Dualplex™ assistants — test GPT Realtime 2 with a typical script. The instruction-following gain is measurable on the very first call.
2. Test Cartesia voices on pronunciation-critical scripts — proper nouns, pharma, legal — Sonic 3.5 should be visibly more consistent.
3. Evaluate GPT-5.5 — anyone currently using GPT-4o or GPT-5 for FAQ bots should run a price/performance comparison. An A/B test on 100 calls is enough to make the call.
4. Use carrier self-service — if you already have Twilio/Telnyx/Zadarma/DIDLogic, you can run the onboarding yourself. Saves wait time and support roundtrips.
5. Integrate AI-pause endpoints into your helpdesk — anyone running hybrid support workflows should wire the new endpoints into escalation routing.

Conclusion

The May 2026 release addresses the three most frequent pain points of our power users: instruction following in multi-step conversations, voice naturalness in long calls, and friction in connecting your own telephony. GPT Realtime 2 plus Cartesia Sonic 3.5 plus self-service SIP wizards meaningfully raise the bar that production voice agents are measured by in 2026.

If you don't yet run Famulor in production, ask for a live demo — a demo configuration takes 20 minutes to build. For existing customers: most upgrades are automatically active; a few (GPT-5.5, auto top-up, AI pause) need to be toggled per assistant or account.

The full Famulor changelog and blog lists all release notes from recent weeks. Questions: support@famulor.io or directly inside the platform via chat widget.

FAQ

Do I have to enable GPT Realtime 2 manually?

No. All assistants in speech-to-speech and Dualplex™ mode now run automatically on GPT Realtime 2. There is no configuration toggle — the upgrade is transparent and backward compatible.

What does GPT-5.5 cost in pipeline mode?

GPT-5.5 is available as a standard LLM option in pipeline mode. Full pricing details are at famulor.io/pricing in the assistant configuration based on your chosen voice and LLM stack.

Which carriers does the new self-service importer support?

Twilio SIP Trunks, Telnyx, Zadarma, and DIDLogic. Additional carriers can be connected on request via custom SIP configuration. The wizard lives in the assistant setup area.

What is different between Cartesia Sonic 3.5 and Sonic 3?

Sonic 3.5 sounds more natural across long calls, is more stable on difficult words and proper nouns, and preserves voice characteristics over the full call duration. The upgrade applies automatically to all Cartesia voices with no configuration change.

How does AI pause per conversation work?

Two new API endpoints — POST /conversation/{id}/ai-disable and POST /conversation/{id}/ai-enable — toggle AI processing per conversation off or on. This is ideal for hybrid workflows where a human takes over temporarily.

Is post-call re-transcription billed separately?

Yes. Re-transcription is transparently logged in the call view of the affected call so you can see per call whether and how often re-transcription ran. Billing follows your plan's standard STT pricing.

Does auto top-up work for voice credits too?

Currently auto top-up has rolled out for chat credits. An analogous feature for voice credits is on the roadmap. Until then, voice credits can be topped up via one-click or subscription.

Is split view available across all plan tiers?

Yes. Split-view detail panels for Calls and Conversations are part of all plan tiers. There is no separate add-on.

How do I migrate from pipeline mode to Dualplex™?

In the assistant editor under "architecture" you can toggle modes. We recommend a brief test run, since Dualplex™ has different conversational dynamics — faster, with true turn-taking. Our support team can help with tuning if needed.

Where can I find the full changelog?

All release notes, including older versions, are on the Famulor blog page as well as in the changelog section of the help center.