Google Joins Rival Tech Giants in Launching Decision Models After TypeSafe's Jev Breakthrough

Google has unveiled its own decision model, following OpenAI and Microsoft, in a race to adopt a new AI paradigm that selects answers from a fixed list with probability scores in milliseconds. TypeSafe AI's Jev, launched in September, has already seen rapid adoption, with 35% of paid Vercel teams using it within two weeks. Despite competition, chatbots still outperform these models on complex decisions, albeit at significantly higher cost.
On November 11, 2026, Google launched Gemini Decision 1 and its Decisions API, joining OpenAI and Microsoft in adopting a new AI paradigm centered on fast, probabilistic decision-making rather than text generation. This shift follows the September 15 launch of TypeSafe AI’s Jev, the first widely adopted decision model, which quickly gained traction—35% of paid Vercel teams used it within two weeks. Jev’s success prompted rapid industry adoption, with decision models capturing 26% of AI requests on Vercel’s gateway by October, despite costing just a fraction of language models. Google’s model, like OpenAI’s and Microsoft’s, charges $0.10 per million input tokens with no output cost, emphasizing speed and low latency. Meanwhile, open alternatives like Liquid AI’s d1-3B and Nace AI’s Drex series are outperforming Jev on public benchmarks, with Drex Max scoring 67.52 on Decision Index 0.4.1 versus Jev’s 60.70. TypeSafe AI, now valued at $7.5 billion after an $870 million funding round led by Andreessen Horowitz, is the category’s originator, with CEO Diogo Almeida previously at OpenAI. Microsoft’s model is built on Alibaba’s Qwen, while Shanghai AI Lab’s 4B Intern-Decision model achieved 90.02% accuracy across seven test suites, outperforming Jev. Despite these advances, chatbots like GPT-6 Luna and Claude Haiku 5.5 still lead in complex decision accuracy—by about 3 points—though at 15 times the cost and 40 times the latency. The market is now bifurcating: generation and judgment are becoming distinct AI workloads, with decision models handling routine, low-cost tasks while language models retain high-value, complex work. What remains unresolved is whether open, cheaper models will displace proprietary ones, or if the cost-performance gap will drive further consolidation.
The rise of decision models could reshape how AI is deployed in production systems, shifting computational load from expensive, slow language models to faster, cheaper alternatives optimized for judgment. This may lower barriers for startups and developers needing rapid, reliable AI decisions without the overhead of full-text generation. As decision models now handle nearly a quarter of AI requests on Vercel’s gateway, their adoption may accelerate cost reductions across software stacks, particularly in automation, content moderation, and workflow routing. However, the dominance of chatbots on complex decisions—despite higher cost and latency—suggests that trust in AI judgment remains fragile, and users may still prefer full-text explanations even when speed is prioritized. The trend echoes the earlier shift from rule-based systems to neural networks, where initial efficiency gains were tempered by opacity and reliability concerns. What may emerge is a hybrid model: fast, probabilistic decisions for routine tasks, paired with human-in-the-loop or full-model verification for high-stakes outcomes. The next phase could involve standardization of decision benchmarks, increased use of private data to prevent gaming, and deeper integration of decision models into infrastructure layers. Watch for whether open models like Drex and d1-3B will disrupt the market or if incumbents will maintain control through ecosystem lock-in.