In partnership with
Good morning, {{first_name | AI enthusiast}}.
A former OpenAI researcher who helped invent the technique behind ChatGPT just launched a very different kind of AI. TypeSafe AI’s new model, Jev, skips conversation entirely — it returns fast, typed decisions that its maker claims can’t hallucinate.
Backed by a fresh $40 million seed round, TypeSafe is betting that most of the world’s AI usage will end up machine-to-machine, not human-to-chatbot. If Jev’s speed and pricing hold up outside the lab, does that make today’s giant reasoning models overkill for half of what businesses actually need AI to do?
Today in AI Brief:
TypeSafe’s hallucination-free Jev model
Google’s cheaper, faster Gemini Live voice AI
China’s Z.ai raises $5 billion for frontier models
Product teams aren’t short on ideas. They’re missing a system.
Jira Product Discovery gives teams one place to capture customer feedback, prioritize ideas with consistent frameworks, and build living roadmaps everyone can align on. And when it’s time to build, those decisions connect directly to delivery in Jira, so everyone can see how the roadmap turns into real work.
ChatGPT Co-Creator Launches an AI That Can’t Hallucinate
In Brief: TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, emerged from stealth with Jev, its first “System One Model,” alongside a $40 million seed round led by DCVC.
The Details:
Jev returns typed decisions with confidence scores instead of text, choosing only from a preset list of outputs — which TypeSafe says makes hallucination mathematically impossible.
It responds in 70–500 milliseconds, 40–200x faster than frontier LLMs, and costs just $0.042 per million input tokens, with structured output tokens free.
Developers can join the early-access waitlist now to test Jev on real automation workflows, from routing tickets to approving transactions.
Take Away:
Almeida co-invented the RLHF techniques that shaped ChatGPT, and now he’s betting the next wave of AI value comes from machines making split-second decisions, not chatting with humans. If Jev’s numbers hold up in production, it opens a much cheaper lane for the “boring” automation work LLMs were never built for.
Google’s Gemini 3.8 Live Undercuts OpenAI on Voice AI Pricing
In Brief: Google rolled out Gemini 3.8 Live and 3.8 Live Extended Thinking, its newest real-time voice models, across the Gemini API, AI Studio, Search Live, Gemini Live, and Workspace.
The Details:
Gemini 3.8 Live charges $0.005 a minute for audio input and $0.018 a minute for output — less than half the price of OpenAI’s rival GPT-Live-1.
The models process visual input in near real time and can automatically detect and switch between 97 supported languages mid-conversation.
The Extended Thinking variant adds multi-step reasoning for harder tasks, and already powers Gmail Live, Docs Live, and Keep Live.
Take Away:
Google is pairing genuinely improved reasoning with a price cut, not just matching rivals on cost. That combination makes voice interfaces cheap enough for developers to bolt onto products that never considered talking to users before.
Hire Ava, the AI BDR built for enterprise
Ava is the first AI BDR to run outbound end to end, and you decide whether she runs autonomously or on copilot.
She finds leads or ingests accounts from your CRM, enriches them, sends personalized emails on behalf of your reps, follows up, handles replies and books meetings. Website visitor de-anonymization, intent signals, and a parallel dialer come built in.
A small team can manage her centrally for thousands of reps who never log in. Everything syncs two-way with Salesforce and HubSpot.
Ava runs outbound for companies like DoorDash and Grammarly, and one customer deploys her across 1,000+ reps. Ava is SOC 2 Type II audited, SSO and GDPR ready. Ava is how revenue teams grow pipeline without growing headcount.
China’s Z.ai Raises $5 Billion to Chase Frontier Models
In Brief: Z.ai, the Hong Kong-listed lab formerly known as Zhipu AI, closed roughly $5 billion through a share placement and convertible bond sale — its second major raise in two months.
The Details:
60% of the proceeds go toward next-generation model R&D, compute infrastructure, and a fully self-training system, according to the filing.
New shares priced at a 10% discount to Z.ai’s closing price, expanding its share count by roughly 10%, with the placement set to close September 16.
The raise follows MiniMax’s HK$16 billion placement and ByteDance’s $29.6 billion loan for AI investment, part of a broader compute-buying spree among Chinese labs.
Take Away:
Z.ai is now stockpiling cash and compute at the same pace as its much larger rivals, signaling that China’s AI race is being funded as aggressively as the model releases suggest. Expect the capital-raising pace among Chinese labs to keep tracking the release calendar in Silicon Valley.
Everything else in AI
Odyssey unveiled Odyssey-3, a foundation “world model” that uses one architecture to control robots, self-driving cars, and drones after learning from just tens of hours of demonstration data.
Salesforce reported that its new Koa reasoning model for Agentforce cuts CRM agent errors 3x versus general-purpose models in early pilots with UChicago Medicine and Formula 1.
Meta launched Meta One, a tiered subscription spanning Instagram, Facebook, and WhatsApp with expanded AI tools, already pulling in 15 million subscriptions and trials.
Google mapped the complete brain of a male fruit fly — 166,000 neurons and 125 million connections — using AI to accelerate progress toward mapping the human brain.