Workers AI gets a speed boost, batch workload support, more LoRAs, new models, and a refreshed dashboard

Cloudflare announces major Workers AI updates: 2–4x faster inference via speculative decoding, prefix caching, and an updated backend; an asynchronous batch API for large, non-latency-sensitive workloads; expanded LoRA support (more models, higher ranks, larger adapters); quality-of-life improvements including updated pricing and a new usage dashboard; and over 10 new or updated models such as @cf/deepseek-ai/deepseek-r1-distill-qwen-32b, @cf/baai/bge-m3, @cf/openai/whisper-large-v3-turbo, @cf/mistralai/mistral-small-3.1-24b-instruct, @cf/google/gemma-3-12b-it, @cf/qwen/qwq-32b, and @cf/qwen/qwen2.5-coder-32b-instruct. These changes target better performance, reliability, customization, and usability for distributed inference at scale.