TypeSafe AI, an AI startup founded by ChatGPT co-inventor, emerges from stealth with $40M to build AI that’s 100X faster and cheaper

TypeSafe AI has emerged from stealth with $40 million in seed funding and an ambitious pitch: the AI models that became great at talking to humans may be the wrong models for running software.

The San Francisco startup was founded in 2024 by Diogo Almeida, a former OpenAI researcher who worked on reinforcement learning from human feedback and early ChatGPT research, alongside Erik Gafni and Sasha Sheng. DCVC led the seed round.

TypeSafe is taking aim at a problem that has become harder to ignore as companies put AI into production. Large frontier models can reason, write code, and carry on remarkably human conversations, yet production software often needs something less conversational and more predictable. A model that changes its approach between requests or confidently invents an answer can create problems when another piece of software is relying on that output to make a decision.

Almeida thinks AI needs another path.

“TypeSafe was founded to pursue an alternative path for AI research, focused on machine-native AI,” said Almeida, TypeSafe’s co-founder and CEO. “I spent years working on models designed to make AI better at interacting with people. But if AI is going to fundamentally change how work gets done, people can’t be the only consumers of intelligence. Most intelligence should eventually live inside software, running quietly in the background. TypeSafe is building for that future.”

From ChatGPT to AI built for machines

That idea sits at the center of TypeSafe’s first model, Jev, named after Jevons Paradox, the economic observation that greater efficiency can sometimes increase total consumption rather than reduce it.

Instead of positioning Jev as another chatbot competing for people’s attention, TypeSafe wants developers to treat it more like a software primitive: a component applications can repeatedly call whenever they need semantic judgment or an intelligent decision.

TypeSafe says Jev can produce frontier-level intelligence with latency below 100 milliseconds and can be up to 100 times faster and less expensive than other frontier models. The company says the model can generate hundreds of outputs in parallel from a single prompt and return calibrated confidence scores, giving developers a way to decide when an application can act on its own and when it should defer.

Those are TypeSafe’s performance claims, and the real test will come as Jev reaches more developers and gets measured against competing models across production workloads.

The startup is currently offering Jev through an early-access waitlist.

Almeida offered an even more aggressive account of the company’s thesis in a post on X:

“After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions

AFAICT the shortest path to AI-based economic revolution”

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x… pic.twitter.com/JSybNG2BKJ

— Diogo Almeida (@CompleteSkeptic) September 15, 2026

TypeSafe AI: A $40 million bet on a different AI architecture

The size of TypeSafe’s seed round makes the launch particularly interesting. DCVC isn’t backing another application layered over APIs from OpenAI, Anthropic or Google. It is financing a different thesis about what AI models themselves should look like once their primary users are machines.

“TypeSafe is approaching one of the biggest remaining challenges in AI: turning increasingly capable models into technology that developers can reliably build into products at scale,” said James Hardiman, General Partner at DCVC. “Diogo, Erik, and Sasha bring the technical depth and conviction to rethink how models are built for software. Their approach to composable intelligence will unlock an entirely new generation of applications.”

A compelling economic argument underpins the idea. If AI becomes cheap and fast enough, developers could insert intelligent decisions into places where calling a large frontier model currently makes little financial or technical sense. Instead of one expensive model handling an entire task, software could make dozens, hundreds or potentially thousands of small AI-assisted decisions behind the scenes.

That is where the Jevons Paradox reference starts to make sense. Lowering the cost of intelligence could lead developers to consume far more of it.

TypeSafe still has plenty to prove. OpenAI, Google, and other major model developers continue pushing inference costs down and speeds up. Any performance advantage Jev has today could come under pressure as larger competitors improve their own models and infrastructure.

Yet TypeSafe is making a different bet from much of the AI industry. Almeida spent years helping teach machines how to interact with humans. His new company is betting that one of AI’s biggest markets may emerge when humans are no longer sitting on the other side of every prompt.