Customization Emerges as Key Trend in Rapidly Growing Decision Model AI Sector

Image: TechCrunch · Source
As decision models quickly gain widespread adoption for efficient AI-driven choices, startups like Sqwish Labs pioneer customization by training models on client data to refine accuracy and relevance.
Decision models, a specialized branch of large language models (LLMs) designed to select from preset options rather than generate open-ended text, have seen explosive growth since their recent introduction. By October 2026, over 25% of AI model requests on Vercel's platform targeted decision models, highlighting the rising demand for fast, cost-effective automated decision-making in production environments.
Unlike traditional LLMs that focus on generating detailed reasoning or responses, decision models prioritize swift, reliable choices with confidence metrics, making them invaluable where latency and cost matter. The Jev family of decision models from TypeSafe set a high adoption bar, reportedly processing over 1.5 billion queries weekly via services like OpenRouter and driving a valuation of $7.5 billion for their creator.
The relative ease of creating decision models has prompted many entrants, including hyperscale cloud providers such as Microsoft, which recently launched its Microsoft-Decision-1.1 model widely on Azure. The proliferation underscores the growing recognition that decision models are critical components in AI stacks.
U.K.-based startup Sqwish Labs, co-founded by Cambridge graduates Ushnish Sengupta and Federica Freddi, distinguishes itself by enabling clients to continually customize decision models with their own production data. This approach extends beyond reliance on synthetic data, which both Sqwish and TypeSafe use extensively, to fine-tune model judgment in real-world settings post-training. Sengupta described this capability as key to avoiding commoditization and delivering ongoing improvements tailored to customer needs.
Benchmarks to compare decision models remain imperfect proxies for real-world performance. Industry leaders caution that high benchmark scores often do not translate directly to enterprise success because constraints like cost, latency, and calibration across unseen use cases play a decisive role. Both Microsoft and TypeSafe emphasize balanced performance over peak benchmark results.
Looking ahead, experts anticipate decision models (dubbed “System One” intelligence) will complement more generalized frontier LLMs (“System Two” intelligence), creating synergistic AI systems. For startups like Sqwish, focusing exclusively on decision models represents a substantial market opportunity distinct from the multitrillion-dollar ambitions of sector giants.
This evolution reflects an emerging frontier in AI where bespoke, data-driven customization underpins practical deployment, meeting the needs of businesses requiring precise, efficient decision-making capabilities rather than solely expansive generative models.
Sources and original reporting
Read the original source ↗

Comments (0)
No comments yet. Start the discussion.
Write a comment
Comments are published after moderation. Your name and comment will be visible publicly. Account