We’ve been thinking a lot about how to best evaluate multimodal models, especially with so much new work happening around voice agents and speech-to-speech benchmarx.
Before putting out benchmarks of our own, we wanted to spend more time understanding the core decisions that go into building a good evaluation in our latest blog post.
We’ve historically shared a lot about the data we build at Mundo, but there’s so much cool work happening internally that we’re excited to start sharing more of :)
come by the lounge!!!
Replying to @hellomundoai
@hellomundoai is headed to Interspeech 2026 (and sponsoring the conference lounge)!
A huge focus of mine right now is building better speech evals/benchmarks that reflect real-world distributions - would love to chat with others working on this!
Excited to catch up with old friends and meet new folks! DM me if you'd like to chat speech data / evals 🫡
Angus Hardy-Francis retweeted
Even though this week feels like the GPT-6 Astra week, there were also many updates on the RealTime AI front! 🎉
On the STT side, two big companies released updates: Meta with Muse Voice Transcribe and Microsoft with MAI-Transcribe-2. Competition is heating up here! On the TTS side, Inworld’s Realtime TTS-2 went GA and topped the Artificial Analysis TTS arena. For turn detection, we have a new contender in Sesame's TurnBench, with Tavus’ Sparrow-2 taking the #1 spot.
LiveKit also dropped multiple updates this week, including Connectors as an alternative to SIP integrations, a staging environment for agents, and a new release of their Agents framework.
On the product side, Google keeps adding voice features, including additions to Google Suite and improvements to Google Translate on Android. Additionally, Tesla upgraded its voice interactions to the latest Grok model, and Genesys partnered with ElevenLabs on agentic voice for contact centers.
More news and updates are in the newsletter 👇. Have a great week!
Angus Hardy-Francis retweeted
Today marks a major step for Mistral: we’re announcing a €3B Series D, the largest equity round ever raised by a European tech company, just three years after launch.
Today, @GoogleDeepMind and @GoogleResearch are introducing WeatherNext 3, our most advanced and accurate global weather AI model to date. It uses real-time satellite data to generate hourly high-resolution forecasts, precise precipitation forecasting, and clean energy variables.
Most AI weather models are trained on data from numerical weather prediction (NWP) models, which come with a six-hour data lag. But by training on real-world observations, WeatherNext 3 is able to bypass these traditional constraints — meaning more people can get more accurate predictions to help them plan ahead.
Starting today, WeatherNext 3 will power forecasts within Search, @GeminiApp, @GoogleMaps, Google Maps Platform Weather API and Google Earth Engine.
Angus Hardy-Francis retweeted
We have processed our first submission to TurnBench.
Congrats to the @tavus team for Sparrow-2, which is a new #1 on the TurnBench leaderboard.
At @hellomundoai, we recently worked with @sesame and others on TurnBench, an open benchmark for measuring turn-taking in spoken dialogue.
One of the hardest challenges we faced: how do you collect specific conversational behaviours without making them feel staged?
Ask someone to “interrupt more” or “sound angry,” and you often end up measuring their performance of the instruction rather than the behaviour itself.
We wrote about what we learned building TurnBench:
mundoai.world/research/the-n…
Angus Hardy-Francis retweeted
We're releasing a 30h hand-annotated benchmark corpus, a 104h training set, a public leaderboard and our full evaluation methodology.
arxiv.org/abs/2608.25218
Thank you to our collaborators at @hellomundoai @otodotearth and researchers at CMU, NTU, Academia Sinica, and Brno!
Angus Hardy-Francis retweeted
I'm excited to share some of what i've been working on at Sesame!
turnbench.sesame.com is our turn-taking benchmark for real-time conversation: when to speak and when to yield
Angus Hardy-Francis retweeted
Learn more ↓
goo.gle/4zEJFFU
Angus Hardy-Francis retweeted
I was today years old when I learned that Slack is an acronym for Searchable Log of All Communication and Knowledge
Angus Hardy-Francis retweeted
How can companies verify age online with confidence, while providing a quick and seamless onboarding experience? #KYC ow.ly/xwSB30oWCPU
Angus Hardy-Francis retweeted
.@FrontFundr upgraded their #onboarding workflow by integrating with #GlobalGateway, our marketplace for identity data and services. Find out how they were able to deliver quick and comfortable experiences for investors by reading our blog post ow.ly/rg7a30oWCPT
Angus Hardy-Francis retweeted
Trulioo will be at this year's Disupting Online Gambling event in London! Drop by our booth to learn more about how we can help you with your identity needs.
ow.ly/kas150uyyHQ
Angus Hardy-Francis retweeted
Thrilled to announce @trulioo’s newest partnership with @Refinitiv @money2020 #money2020EU For the full deets —-> finextra.com/pressarticle/78… @StephenUfford
Angus Hardy-Francis retweeted
Have you seen our latest #GetVerified parody music video? Trulioo’s third annual parody music video, it chronicles the journey of a young female entrepreneur establishing her business from the ground up. Enjoy!ow.ly/EXQf50uvNIG
Angus Hardy-Francis retweeted
.@StephenUfford takes PYMNTS inside how @trulioo's team up with @Refinitiv aims to boost digital ID and financial inclusion: pymnts.com/financial-inclusi…
Angus Hardy-Francis retweeted
Catch our CEO @StephenUfford at #M2020EU for a fireside chat on #financialinclusion and digital identity, and how the two are inextricably linked. The session takes place on the Tech Combustion Stage at 5:45 PM (CEST). @money2020
Angus Hardy-Francis retweeted
The #Trulicrew sends our support for Max @foshma and #TeamMobile! 🏁🏃🏻🏁
#RTWrace #PaymentsRace #M2020EU @money2020






















