TypeSafe raises $870M at $7.5B valuation for non-text AI model Jev
TypeSafe AI, the developer of Jev, a new type of AI model that launched on September 15, has raised $870 million at a $7.5 billion valuation. The round was led by Andreessen Horowitz with participation from Sequoia and existing investor DCVC. The fundraise comes weeks after Jev went viral and the startup claimed a third of Fortune 500 companies are already using the model.
Jev outputs probabilities not text
Jev is built on a transformer architecture but it is not a large language model. It does not output text. Instead it produces probabilities, what the company calls calibrated decisions. TypeSafe claims Jev works significantly faster and uses far fewer tokens than LLMs. The company positions its approach as uniquely suited for automating tasks rather than generating text or code.
Founders from OpenAI and Meta
Diogo Almeida, a former OpenAI researcher, co-founded TypeSafe in 2024 with Sasha Sheng, a former Meta research engineer, and Erik Gafni, an engineer and entrepreneur. Almeida said the industry has been super good at human language for four years but that capability is not useful for automation because computers speak a different language.
What calibrated decisions mean for automation
Traditional LLMs generate text or code. Jev produces probability distributions that can drive automated workflows directly. This removes the parsing layer that turns model text into machine actions. For tasks like routing, filtering, or real-time classification the model returns a decision vector instead of a sentence. TypeSafe argues this matches how computers actually operate and eliminates the token overhead of natural language generation.
Token efficiency and speed claims
The company says Jev uses far fewer tokens than comparable LLMs. If verified this could translate to lower inference costs at high throughput. Engineers evaluating Jev should benchmark against their specific workloads since the model's training data and architectural constraints differ from standard LLMs. The output format — probabilities rather than text — requires different integration patterns than LLM pipelines.
Investor confidence in non-text AI stack
The $870 million round at a $7.5 billion valuation signals that venture capital sees value in a non-text AI stack that can operate across enterprise operations. With Andreessen Horowitz leading and Sequoia and DCVC participating the round reflects confidence in a model that addresses a perceived gap between general-purpose language models and domain-specific automation tools.
Implications for teams building with AI
Teams that previously relied on LLM-generated text may need parsers that consume decision distributions instead. The claim of token efficiency could make Jev attractive for high-throughput use cases where every inference call counts toward operational cost. Engineers should validate the speed and token claims against production workloads before committing to the new integration pattern.