Understand this story
On November 17, 2026, TypeSafe AI released a decision model called Jev, which it claims does not produce hallucinations. According to IEEE Spectrum, the model claims end-to-end latency of 70 to 500 milliseconds, with pricing at $0.042 per million input tokens and free output. A preprint test from Carnegie Mellon University showed that Jev's median latency was 150 milliseconds, compared with 1.89 seconds for GPT-6 Astra. The researchers noted that the decision model eliminates output hallucinations, but it can still make errors in mathematics, logic, and specialized fields such as medicine, finance, and law; a hybrid approach that hands low-confidence results to an LLM for review slightly outperformed using GPT-6 alone in the tests, at 41% of the latter's cost. The report also said that Cloudflare, Amazon, OpenAI, and Microsoft subsequently launched similar models.
AI-generated from public reports
Latest development · Report
TypeSafe AI's Jev popularizes decision models, with Cloudflare and other companies following suitDevelopments
1 confirmed developments
- Source:IEEE Spectrum
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 17
IEEE SpectrumSelected
TypeSafe AI's Jev popularizes decision models, with Cloudflare and other companies following suit
TypeSafe AI released a decision model called Jev that it claims does not produce hallucinations, and Cloudflare, Amazon, OpenAI, and Microsoft subsequently launched similar models. Jev claims an end-to-end latency of 70 to 500 milliseconds, with pricing at $0.042 per million input tokens and free output; a preprint test from Carnegie Mellon University showed a median latency of 150 milliseconds, compared with 1.89 seconds for GPT-6 Astra. Researchers noted that decision models eliminate output hallucinations but still make errors in mathematics, logic, and specialized fields such as medicine, finance, and law; a hybrid approach that hands low-confidence results to an LLM for review performed slightly better than using GPT-6 alone in the tests, at 41% of the latter's cost.
AI Trends