TypeSafe AI Releases Jev Decision Model, Claiming Lower Latency and Reduced Hallucinations
- Published
- Source
- IEEESpectrumAI
On 15 September, TypeSafe AI released Jev, a model the startup calls a “System One model.” TypeSafe AI claims Jev is cheaper and faster than large language models (LLMs) and cannot hallucinate. The company recently raised $870 million at a $7.5 billion valuation. Jev’s release has been followed by decision models from Cloudflare, Amazon, OpenAI, and Microsoft, as well as dozens of models from startups, researchers, and independent developers. TypeSafe AI’s claims about Jev’s capabilities remain company claims; the article also cites independent testing of its latency.
Decision models return a constrained choice, often with probabilities, rather than generating a sequence of text tokens. A preprint co-authored by Carnegie Mellon University professor Ramayya Krishnan reported Jev’s median latency at 150 milliseconds across decision-accuracy benchmarks, compared with 1.89 seconds for OpenAI’s GPT-6 Astra. TypeSafe AI lists input tokens at $0.042 per million and says output tokens are free. The approach could suit high-volume decisions such as content moderation, while retaining broader task applicability than traditional, domain-specific classifiers. Cloudflare’s Clef illustrates the approach: it is an open-weights model based on Alibaba’s Qwen.
TypeSafe AI has disclosed few details about how Jev achieves its speed, leaving its technical implementation an unresolved point. The article describes decision models as requiring a single forward pass and producing structured outputs, but that explanation is given for Cloudflare’s Clef, not Jev. Further evidence on Jev’s performance beyond the cited benchmarks, and on how broadly decision models generalize across tasks, remains a monitoring point.
Original report
AI Companies Develop Decision Models to Avoid Hallucinations
IEEESpectrumAI · Matthew S. SmithThis briefing is an independently written summary based on publicly available reporting and is provided for industry information and news discovery. The original report and source publication are credited and linked where applicable. RobotToday does not claim ownership of third-party source material.
Rights concerns? Contact [email protected] with the relevant URL and details. We will review the matter and take appropriate action where warranted.
More related news
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
Related Suppliers
OpenAI
US AI research lab and product company founded 2015; develops GPT, DALL-E, and robotics/embodied AI foundation models targeting physical-world AGI.
Robocurve
San Francisco public benefit corporation founded 2026 that benchmarks frontier AI models on real robot arms and maintains the Inspect Robots framework.
Shenzhen Liangyuan Innovation Technology Co., Ltd. 深圳亮源新创科技有限公司
Shenzhen embodied-AI lab led by ex-OpenAI researcher Roger Jiang, building Physical AI foundation models Light-O1, Light REACT and LightNav-0
I Hub Research and Robotics Private Limited (iHub Robotics)
Kochi startup (relaunched 2022) making the TARA Gen1 semi-humanoid service robot and Daksha industrial humanoid, exported to the UAE and Saudi Arabia.
Archetype AI
Physical AI platform company deploying the Newton foundation model to fuse sensor data with language for real-world industrial intelligence.
AI Robot Association (AIRoA)
Tokyo non-profit association of Japanese industry and academia collecting open robot data and building robot foundation models, backed by NEDO.
AI, Data and Robotics Association (Adra, asbl)
European public-private partnership association driving AI, data, and robotics research and innovation under Horizon Europe.
FMC³ Robotics
Ingolstadt, Germany embodied-AI firm whose RoboBrain software and DATOS haptic-glove kit run on partner robots; MoU with Vector and PEM Motion in 2026.
Synthium
Synthium is a Singapore-based, Antler-backed startup that crowdsources gamified simulations to generate training data for humanoid robot AI models.
Joby Aviation
Santa Cruz, CA eVTOL developer (NYSE: JOBY); S4 air taxi seats 5, flies 200 mph up to 100 miles; backed by Toyota and Delta Air Lines; founded 2009.
Robust.AI
US developer of AI-powered collaborative mobile robots and fleet software for warehouse automation, known for the Carter AMR.
Peridio
Peridio makes Avocado OS, an immutable embedded Linux platform with built-in OTA fleet management, for robotics and physical AI hardware makers.
Robotics needs a service framework.
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
Join the RobotToday community on LinkedIn
Daily robotics news, in-depth analysis, conference highlights, and discussions with professionals worldwide.