Voice models ship while developer infrastructure quietly hardens
Gemini 3.1 Flash Live, Voxtral TTS and ChatGPT product discovery, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
Google DeepMind
Gemini 3.1 Flash Live
DEPLOYif you build voice agents
WHAT SHIPPEDGoogle's highest-quality audio model, with ComplexFuncBench Audio at 90.8% and Scale AI Audio MultiChallenge at 36.1% with thinking on. Gemini Live follows conversation for twice as long and Search Live expands to over 200 countries.
THE CATCHAll audio is watermarked with SynthID, which matters for compliance teams but adds a detection layer you should test before production.
IF YOU RUNcustomer support voice agents that need to handle interruptions and multi-step tasks
SOURCEGoogle DeepMind, Gemini 3.1 Flash Live: Making audio AI more natural and reliable
Mistral
Voxtral TTS
DEPLOYif you run voice workflows in Europe
WHAT SHIPPEDA 4B-parameter text-to-speech model with 70ms latency for a 10-second sample, 9 languages, and voice adaptation from a 3-second reference. Priced at $0.016 per 1k characters.
THE CATCHHuman evaluation shows Voxtral beats ElevenLabs Flash v2.5 on naturalness at 68.4% win rate, but the model is built on Ministral 3B, so check whether your stack already assumes a different TTS provider.
IF YOU RUNcustomer support or compliance voice pipelines that need multilingual output
SOURCEMistral, Speaking of Voxtral
OpenAI
ChatGPT product discovery
WATCHif you sell through Shopify or retail partners
WHAT SHIPPEDVisual product browsing, side-by-side comparison, and image-based search rolling out to all ChatGPT free, Go, Plus, and Pro users this week. Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair are integrated.
THE CATCHOpenAI is dropping Instant Checkout in favor of merchant-owned checkout, so the conversion path now depends on each retailer's own flow rather than a native ChatGPT purchase.
IF YOU RUNa Shopify store and want product data surfaced in ChatGPT conversations
SOURCEOpenAI, Powering Product Discovery in ChatGPT
The Deploy Log
Models and APIs
| DEPLOYif you generate music for video or marketing Lyria 3 Pro. Lyria 3 Pro creates tracks up to 3 minutes long with intros, verses, choruses, and bridges, available in Vertex AI, Google AI Studio, the Gemini API, Google Vids, the Gemini app, and ProducerAI. Google DeepMind |
Tools and products
| DEPLOYif your team codes with Codex CLI Vercel plugin for Codex. The Vercel plugin now supports OpenAI Codex and Codex CLI, giving teams access to over 39 platform skills, three specialist agents, and real-time code validation. Vercel |
| WATCHif your team lives in Notion Notion 3.4. Notion 3.4 adds dashboard view for databases, a redesigned opt-in sidebar, presentation mode in beta for Plus, Business, and Enterprise, a tabs block, image generation, and page archiving. Notion |
| DEPLOYif you reconcile AI spend across providers AI Gateway Custom Reporting. The Custom Reporting API for AI Gateway is in beta for Pro and Enterprise plans, giving programmatic access to cost, token usage, and request volume across AI Gateway traffic including BYOK requests, with one platform saving over $80K by replacing a third-party proxy. Vercel |
Left out
- Archive pages: the publisher's page does not show it available yet
- Chat SDK scheduled Slack: the publisher's page does not show it available yet
- Team default build machine: the publisher's page does not show it available yet
- Named sandboxes: it did not carry a number that moved
- Runtime log error codes: it did not carry a number that moved
- Activity Log in CLI: it did not carry a number that moved
- Chat SDK concurrency: it did not carry a number that moved
- Elastic build machines: the condition it attaches to the call did not hold up
- Persistent sandboxes: it did not carry a number that moved
- Teen safety policies: it did not carry a number that moved
The pattern
Voice interfaces are shipping as consumer-ready products while the infrastructure underneath them quietly hardens. Gemini 3.1 Flash Live, Voxtral TTS, and Lyria 3 Pro all push audio generation forward, but each carries a compliance or integration constraint: SynthID watermarking, a Ministral 3B dependency, and distribution across Vertex AI, Google AI Studio, the Gemini API, Google Vids, the Gemini app, and ProducerAI. Meanwhile, Vercel's plugin for Codex and Codex CLI, Notion 3.4, and the AI Gateway Custom Reporting API in beta for Pro and Enterprise plans are not flashy launches. They are the rails that let teams deploy those voice models, track cost and token usage across BYOK requests, and replace a third-party proxy to save over $80K.
The product discovery shift in ChatGPT points the same direction. Visual browsing, side-by-side comparison, and image-based search reach all ChatGPT free, Go, Plus, and Pro users this week, with Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair integrated. But OpenAI drops Instant Checkout in favor of merchant-owned checkout, so the conversion path now depends on each retailer's own flow. That is the same tradeoff as Voxtral at $0.016 per 1k characters beating ElevenLabs Flash v2.5 on naturalness at 68.4% win rate: the model is impressive, but the surrounding system, not the model, decides whether it works in production.