Another AI startup has emerged in Silicon Valley that has the industry buzzing with excitement and capital rushing to place bets. Reports indicate that TypeSafe AI, the startup behind the new model Jev, has already begun financing discussions with investors, planning to complete a massive funding round reaching $1 billion or even more; unnamed investors have already expressed valuation intentions of $10 billion and above. Just last week, the company had only announced the completion of a $40 million funding round, which at the time carried a valuation of merely $200 million according to PitchBook data. Negotiations are still at an early stage, and the situation could change significantly in the coming days to weeks. But the market enthusiasm is very real: at an AI summit held in San Francisco on Wednesday, at least two speakers mentioned Jev and gave it extremely high praise. Dion Harris, Senior Director of High-Performance Computing at Nvidia, described it as "shockingly fast," while also noting that the model still has limitations. The explosive market interest has also driven up TypeSafe's computing costs, pushing the company toward completing a new funding round as quickly as possible. A nearly identical situation previously occurred with AI agent startup Instinct. If TypeSafe can complete this funding round at such a high valuation, it would demonstrate that even after a wave of new AI labs founded by former Google DeepMind and OpenAI employees emerged over the past two years, investors remain willing to make big bets on startups that bear high training costs to develop entirely new AI models. The company was co-founded two years ago by a former OpenAI researcher, and its Jev model is specifically designed to output numerical results and probability confidence levels. Unlike the large models from OpenAI and Anthropic that generate text sentences word by word, Jev's outputs come with the model's confidence level for each answer. Venture capital firms are excited about it because the model was designed from the ground up to interact with AI agents, rather than with human users like existing foundation models. The Jev development team says the architecture is significantly cheaper and substantially faster than traditional large language models. TypeSafe claims that in some scenarios, the model can be nearly 200 times faster than leading frontier large models while costing only one four-hundredth as much. A company spokesperson declined to comment. Some users say they only use Jev for very specific scenarios: for example, email classification, agent status monitoring, and model task routing. Tamar Yehoshua, Chief Product and AI Officer at productivity software maker Atlassian, said at Wednesday's summit: "Jev is extremely low-cost and extremely fast, but its accuracy falls short of some top-tier large language models." She explained that Atlassian is using it for various classification tasks rather than generating large blocks of text. "In a large number of business scenarios, we not only use large language models, but there are also many scenarios where we would never have used large models before — because the cost was too high and the speed too slow, and Jev happens to fill those gaps." TypeSafe was co-founded by former OpenAI researcher Diogo Almeida, former Meta research engineer Sasha Sheng, and entrepreneur Eric Gafni, with the seed round led by DCVC. A batch of startups is now exploring entirely new architectures distinct from traditional large language models, and TypeSafe is one of them. Similar companies include: one founded earlier this year by former OpenAI senior researcher Jerry Tworek, aiming to build models that can continue learning even after training ends; and Inception, which draws on ideas from image generation AI to develop text generation models. Another peer, Core Automation, has been making very smooth progress in fundraising. People familiar with the matter revealed that this startup, founded in January of this year, has already accumulated $630 million in funding, with its latest post-money valuation reaching $3.5 billion. At the summit, Jerry Tworek said he chose to start his own company partly because he judged that going independent would give him access to more computing resources to support his experimental research compared to working at a large corporation. However, Tworek also mentioned that if an investor or a major AI company recognizes Core Automation's achievements and is willing to provide massive GPU resources to fund the project, he would be open to such collaboration as long as their values align.