Resona

🎙️ Introducing Resona. A production-ready speech‑to‑speech API for voice agents you can finally trust. resona.dev Most voice models break in the same places: • overlap + barge‑in • tool calls accuracy • session duration constraints We built Resona to stop losing these fights. ⚡️ Full‑duplex, bi‑directional streaming over WebRTC or WebSocket. Low latency, with natural thinking pauses. Resona responds between 1–3 seconds depending on how much it needs to think — fast enough to keep flow, paced enough to feel human. 🧰 Industry leading tool use accuracy. The difference between “cool demo” and “it saved my ops team.” ComplexFuncBench Audio: 84%. Runner‑up: Google Live API at 71.5%. That margin shows up when workflows get messy. 🧠 Reasoning in audio shouldn’t be a downgrade. BigBench Audio: 100% (saturated). If the model can’t think through sound, the voice is just decoration. 🗣️ Multi‑turn is where agents quietly fail. MultiChallenge (multi‑turn conversation eval): 69%. Second best: OpenAI GPT Realtime (Aug 2025) at 30.5%. Staying coherent across turns is the whole product. 🌍 12 languages, zero config. Native, cross‑lingual agents — no spaghetti of language pipelines. Build once, ship globally. 🕒 Sessions up to 120 minutes. Long calls shouldn’t be a restart loop. Onboarding, troubleshooting, interviews — the stuff that actually takes time. 🔒 Absolute zero data retention Resona says you retain full ownership of voice/audio data, processed only to provide the service All the data (transcripts + audio) is gone the same second a session ends Enterprise options also include dedicated, fully isolated clusters. 👇 Try right now. Free minutes, instant sign up, start talking to Resona within the next 60 seconds. resona.dev/playground 💳 Love it? ship in an afternoon Docs: docs.resona.dev

It's Codex Mobile day. It honestly took longer for this to happen than we expected considering the speed at which the labs have been shipping over the last year.

@jichiep Keep pushing. Hope your incredible work gets more attention soon.

LocalVQE v1.2 released: Increased echo delay range from 500ms to 1 second and fixed over suppression of voices that sound close to noise. I'll just repost the video of v1.1, you can try out a live demo in the link below to compare versions

@RealEverNever @chatgpt21 Missed that if true. Can't recall OpenAI saying that.

@resona_dev @chatgpt21 Nope. OpenAI said multiple times now that they wont release any mini or nano models in the future.

5.6 baby lets go

@argofowl 5.5 mini and nano. Not 5.6

gpt 5.6 is coming sooner than we think 👀

Very unlikely to be new models. Looks like GPT-5.5 mini and nano are around the corner.

@synthwavedd Wooot? OpenAI already onto another model?

ember-alpha beacon-alpha

@jichiep Incredible work already. Followed to keep ourselves in the loop. Keep pushing.

Yes, with the caveat that there are gaps in the training data that need to be filled, but in those kinds of cases classic AEC usually doesn't work either. In any case you can try it with your data on the HF space before investing time on integration. If you find cases where it doesn't work we'll try to fix that in future releases.

LocalVQE v1.1 is out: A tiny ~1M parameter model that does echo supression and noise cancellation in realtime. V1.1 comes with big improvements to audio quality over V1.

@jichiep Really appreciate the great work. Keep going.

GPT 6

@OpenAI That's GPT 6 for you.

OpenAI DevDay is back. San Francisco September 29

Hey @thsottiaux , we need Codex on the go. Like properly. Not the current half baked cloud version. Please.

@theo Nope. Not anymore. Other labs have gone so far ahead that if xAI is to achieve a breakthrough, the other labs are just many times more likely to do so and earlier.

I legitimately believe xAI might have a crazy comeback

@theo @JetBlue Common bro. You manage to find something to whine about every other day.

Spent over $10k flying my team out on @JetBlue and they didn’t bother putting our bags on the plane. They had our bags over an hour before boarding. I recommend flying literally any other airline.

"A ton of new Codex features" are also dropping today 👀

@sama 82.7% on tbench 2.0, Wow. We should keep Sam drunk every now and then, or we won't be getting another week like this any time soon.

GPT-5.5 is here! We hope it's useful to you. I personally like it.

Can this day get any better? ☺️

@soniox_ai Huge congrats Soniox!!! The day can't get any better. GPT 5.5 and Soniox TTS. I have to pinch myself. 🚀

Big moment for Soniox: Today we’re launching Soniox Text-to-Speech. This is a major step forward for us. Soniox started with speech-to-text. Now, with both STT and TTS, we are becoming the voice platform for every language. Soniox TTS is built for the hardest parts of speech generation: - Native-speaker-quality speech in 60+ languages - Hallucination-free speech generation - Alphanumerics spoken correctly like numbers, IDs, addresses - Correct pronunciation for names and foreign words - Ultra-low-latency streaming for real-time voice applications And the pricing is simple: $0.70 per hour of generated speech. What excites us most is the bigger picture: Developers and companies can now work with one provider for the core voice stack: speech-to-text, text-to-speech, multilingual voice, real-time infrastructure, regional deployments, and compliance. This is a big step in our transition from an STT provider to the voice platform for every language. Voice is becoming a core interface for software. But to work globally, it has to be fast, accurate, robust, and affordable across every language. That is what we are building at Soniox. Read the blog post: soniox.com/blog/soniox-te…

GPT 5.5 (Spud) benchmarks are out! It's finally released.