hotwirednews Get GrokBot

Week 15

6–12 Apr 20262026-W15 · 14 filings · 13 companies · 4 net-new

The Orchestrate layer led the week with 4 of 14 filings, from Daily / Pipecat, LiveKit, VideoSDK and Sierra.

Hearspeech-to-text3

GladiaModel providermodel release2026-04-10

Gladia updates Solaria-1 for Hebrew transcription

A language-specific Solaria-1 update gives Gladia a Hebrew path alongside its wider STT coverage; the three-times figure is Gladia’s own claim from 10 April.

Details2 sources · 1 X post

Gladia says Solaria-1 now delivers three-times-more-accurate Hebrew transcription, powered by the open-source Crowd-Transcribe-V5 dataset.

Solaria-1 Hebrew transcription

Hebrew transcription just got a serious upgrade! It is now 3x more accurate, powered by Solaria-1 — trained on ivrit-ai's Crowd-Transcribe-V5, a comprehensive open-source Hebrew language dataset. Try it for yourself (link in comments)👇

https://x.com/gladia_io/status/2042573315177652317
Bland AIModel providermodel release2026-04-09

Bland Fluent multilingual STT claims 5.9% English WER with more patient endpointing

With Fluent Bland runs its own hearing layer as well as TTS (2026-03-11-bland-speech-standalone-tts); the roughly 27% error cut is against an unnamed rival on Bland's own 250-hour set, and the June noise-cancellation V2 filter (2026-06-11-bland-noise-cancellation-v2) sits in front of it.

Also Orchestrate

Details2 sources

Fluent is Bland's in-house multilingual speech-to-text model for its voice agents, recommended over the older Babel and Auto language options.

Fluent (multilingual STT)

Reson8Model providermodel release2026-04-08

Reson8 launches Resonant-1 and Resonant-1-flash speech-to-text

A 6.51% short-form English average and 1,438× realtime flash inference give Reson8 a first model-line claim in April, but the leaderboard figures and the Resonate table labels are the company's own presentation.

Details1 source

Reson8 launched Resonant-1 and Resonant-1-flash, two speech-to-text models it says top the Open ASR Leaderboard for short-form English, with the flash variant transcribing an hour of audio in under three seconds.

Resonant-1 and Resonant-1-flash

Speaktext-to-speech2

LiveKitPlatformlaunch2026-04-08

LiveKit Agents 1.5.2 adds Baseten streaming TTS and Mistral Voxtral TTS

One speak row for Baseten and Voxtral TTS.

Details1 source

livekit-agents 1.5.2 adds WebSocket streaming on Baseten TTS and Mistral Voxtral TTS with ref_audio zero-shot cloning.

livekit-agents 1.5.2

RimeModel providermodel release2026-04-06net-new

Rime ships Mist v3 enterprise TTS with ~40ms TTFB

Keeps Mist voices while rewriting inference for ~40ms p90 TTFB on L40S/RTX 6000 — latency quotes from Trillet and Attune are customer-claimed.

Details1 source

Rime released Mist v3, a high-throughput text-to-speech inference update that keeps existing Mist voices while claiming about 40ms time-to-first-byte (p90) on L40S or RTX 6000 GPUs for enterprise concurrent voice-agent workloads.

Mist v3

Showavatars, video1

LemonSliceModel providermodel release2026-04-10

LemonSlice 2.1 opens a research preview with live gesture control

Control-surface predecessor to 2.1 Flash (17 April) and, later, CWM-1. Not a second row for the Flash latency paper.

Details1 source · 1 X post

LemonSlice 2.1 is a free research-preview interactive avatar model with an API, described as able to wave, show anger, dance, and otherwise act during a real-time call.

LemonSlice 2.1

Introducing LemonSlice-2.1: the most advanced interactive avatars ever built Make your avatars do anything in real-time… wave, get angry, dance and more! Try the FREE research preview NOW And build avatars into your products using our API First 100 devs get $100 in credits 👇

https://x.com/LemonSliceAI/status/2042648703694172329

Orchestrateframeworks, evals4

Daily / PipecatPlatformAPI2026-04-08

Pipecat Context Hub 0.0.16 indexes the framework at a chosen git tag

Users pinned to an older pipecat-ai get search_api results that match their installed API; the notes describe a single-version replacement model, so only one framework version is indexed at a time.

Details2 sources

Pipecat Context Hub 0.0.16 adds refresh --framework-version so the framework can be indexed at a specific git tag, and 0.0.15 added a pipecat_version argument that scores results for compatibility.

Pipecat Context Hub 0.0.16

LiveKitPlatformlaunch2026-04-08

LiveKit Agents 1.5.2 reuses STT across handoffs and adds AsyncToolset

Handoff continuity is the orchestration headline.

Details1 source

livekit-agents 1.5.2 reuses the STT connection across agent handoffs and adds AsyncToolset plus Vertex RAG retrieval as a provider tool.

livekit-agents 1.5.2

VideoSDKPlatformlaunch2026-04-07net-new

VideoSDK ships Prism Agents V1.0.0 with unified cascade/realtime/hybrid Pipeline

Unified Pipeline unlocks hybrid STT-plus-realtime and custom-TTS-on-realtime paths that v0.x blocked — breaking migration required.

Also ThinkHearSpeak

Details1 source

VideoSDK released Prism (Agents V1.0.0), a breaking rearchitecture that replaces separate cascading and realtime pipelines with one Pipeline class that auto-selects cascade, realtime, or hybrid mode, plus decorator hooks and built-in OpenTelemetry observability.

Prism / Agents V1.0.0

SierraPlatformlaunch2026-04-07

Sierra launches Level 1 PCI-compliant payments for AI agents

Secure payment completion inside the conversation removes the IVR handoff at the moment of payment; Sierra says the capability is Level 1 PCI-compliant and Visa-registry verified.

Details2 sources · 1 X post

Sierra launched a payment capability that lets AI agents collect card and ACH payments inside chat or voice conversations and confirm success before the conversation ends.

Level 1 PCI-compliant payments for AI agents

Introducing the first Level 1 PCI-compliant payment capability for AI agents. No more “please hold” or clunky IVRs when it’s time to pay. With Sierra, customers complete secure payments right inside a chat or voice conversation. Read more here: https://t.co/gI0IF55UHH

https://x.com/SierraPlatform/status/2041602967800000881

Connecttelephony, channels1

VobizPlatformlaunch2026-04-08net-new

Vobiz makes TRAI 1600-series numbers available for BFSI

Puts a regulated Indian BFSI caller-ID series on a voice-AI telephony platform, with the company's own activation figure of about seven days rather than a new speech model.

Also Use

Details2 sources · 1 X post

Vobiz said 1600-series numbers are live for BFSI, with activation it puts at about seven days, alongside provisioning, SIP, and scaling support.

TRAI 1600-series numbers

1600-series numbers now live for BFSI!

https://x.com/Vobiz_Ai/status/2041736461826036198

Useagents in market3

HippoEnterpriselaunch2026-04-08net-new

Hippo rolls out Clara, a 24/7 voice agent for homeowners' first notice of loss

A listed carrier (NYSE: HIPO) running its own FNOL voice agent predates Liberate's Guidewire FNOL ship (2026-07-14-liberate-voice-fnol-guidewire) by three months, though the 70% digital-filing figure is Hippo's expectation rather than a measured rate.

Details2 sources

Hippo Holdings put a conversational AI agent named Clara at the front of homeowners' claim intake, capturing and structuring claim details in real time and routing them onward.

Clara from Claims

Cerence AIPlatformpartnership2026-04-08

Cerence expands its BYD partnership with an xUI and CaLLM assistant for the ATTO 2 DM-i

The ATTO 2 DM-i go-live is described as 'this spring', so 8 April is an announcement date, not a launch date. Hyundai Motor India's SoundHound assistant (2026-10-12-hyundai-motor-india-bayon-soundhound) is the dated counterpart from a different vendor, and the two releases are not linked.

Also Think

Details1 source

BYD's new in-car assistant, built on Cerence xUI and the CaLLM language models, supports multi-step conversations, live information, and several languages, and goes live in the ATTO 2 DM-i this spring before other global models.

Cerence xUI and CaLLM for BYD

NumanacPlatformlaunch2026-04-07

Numanac ships iOS app with voice logging, offline-first capture, and a free plan

Moves Numanac from Jan’s public web release into a phone-pocket field logger with a free tier — material channel expansion for growers who will not type forms in the field.

Details3 sources

Numanac released its field iOS app (voice logging, offline-first sync, Alma record assistant) on the App Store and opened a no-card Free plan for grower adoption.

Numanac iOS app