Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Vercel data shows TypeSafe AI's Jev decision model reached 35% of paid teams in two weeks, per TechCrunch
Vercel published AI Gateway data showing that within two weeks of limited release, TypeSafe AI's Jev decision model was used by 35% of paid teams—nearly double the next model, Sonnet 5.5—and was the fastest first-two-week adoption Vercel has recorded. Jev is a lightweight classifier over a few predetermined outputs; Vercel itself replaced GPT-6 Luna with it for harmful-command classification. The same report says open-source models now account for more than 70% of gateway tokens and nearly 25% of spend, while frontier-lab spend has grown tenfold.
- TypeSafe AI releases Jev 1.14 with experimental image input, a Box type and opt-in 262k context
TypeSafe AI released Jev 1.14.0 under the jev-preview alias, describing small improvements across the board with the largest expected gain in multilingual support; jev-latest is set to move to 1.14 in about a week while jev-1.13.0 stays available. The same post opens opt-in danger-zone features: composable PNG/JPEG image input (also on 1.13), a new Box question type that returns one normalized bounding box per question (1.14 only), and a long-context header that raises limits to 262k tokens. TypeSafe says none of the three are production-ready yet; pricing stays $0.042 per million input tokens with free output.
- Jeff-Qwen3.5-0.8B v1.2 ships with 9 LoRA adapters to route big models at 38x the speed
A community release pairs Jeff-Qwen3.5-0.8B v1.2 with 9 LoRA adapters, positioning the 0.8B decision model in front of larger models like Qwen3.8-27B — the post claims 38x faster decisions with 8.7 points more accuracy for under 2 GB of extra memory. Third-party indexes corroborate the model's existence: the Jevland ecosystem index lists it as a local Jev alternative, and the HostYourAI catalog carries a Jeff/GLiFormer 400M variant used as an EU-hosted router.
- Open-source vllm-jev cuts System One inference from 600ms to about 50ms
The open-source vllm-jev engine cut System One decision inference from roughly 600ms to about 50ms, with official benchmarks commonly showing 5x or better.
- MarkTechPost verifies 20 agentic use cases of TypeSafe AI's Jev, per MarkTechPost
MarkTechPost reports Jev, which skips text generation and returns typed decisions with calibrated probabilities, costs $0.042 per million input tokens with free output, and it verified 20 agentic use cases from model routing to tool-call gating.