Cloudflare Open Sources clef and clef-flash Decision Models on Workers AI
Cloudflare open sourced clef and clef-flash, its first homegrown models, now available on Workers AI with a demo and leaderboard.
Global AI Digest
-
-
AutoTrust AI Introduces JEV-27B-VL Open-Weight Multimodal Decision Model
AutoTrust AI introduces JEV-27B-VL, claimed as the first open-weight near-SOTA multimodal decision model, which converts visual state into calibrated action pro…
-
Hugging Face Partners to Fund Open Source Software Behind Open Science Models
Hugging Face is teaming up with OS for Science to fund the open source software stack that open science models depend on.
-
huggingface_hub v2.1.0 Adds Job Controls, ZeroGPU Quota Tracking, Faster Downloads
huggingface_hub v2.1.0 ships job retry/rerun/reschedule controls, ZeroGPU quota tracking, up to 17× faster HfFileSystem downloads via hf_xet, and Inference Endp…
-
Cloudflare Releases Jev Alternative Open Model on Hugging Face
Cloudflare has released an alternative to Jev as an Apache 2.0 licensed model on Hugging Face.
-
Ling 3.1 Flash Free Model Limit Raised to 300 Requests Per Day
The free Ling 3.1 Flash model has been updated, with daily request limits increased to 300, giving developers more free tokens to build with.
-
Cloudflare Releases Clef: Two Open-Sourced Decision Models Topping Quality and Latency Benchmarks
Cloudflare's Workers AI team introduces Clef, their first trained models: two fast and accurate decision models that top benchmarks for quality and latency, rel…
-
OpenRouter Security Center Adds Key Risk Filtering and IP Allowlisting
OpenRouter's Security Center introduces a safety tab for cross-workspace API key management with risk filtering, limits, bulk actions, plus admin IP allowlistin…
-
Microsoft's MAI-Voice TTS Models Now Available on OpenRouter
OpenRouter now offers Microsoft's MAI-Voice-2.1-Flash (low-latency, $15/1M chars) and MAI-Voice-2.1 (highest fidelity, $22/1M chars) text-to-speech models, both…
-
Same Claude Model Shows 13x Latency Gap Between Vertex and Bedrock Endpoints
Measured p50 latency for claude-opus-5-5 varies dramatically across endpoints, with Vertex US at 1.32s versus Bedrock at 17.9s, and the fastest endpoint shiftin…
