Meta Llama 3 available on Cloudflare Workers AI

Cloudflare announces support for Meta Llama 3 8B (including the instruct model) on Workers AI, emphasizing simple, cost-effective serverless inference for developers. The post highlights performance gains on benchmarks (MMLU, GPQA, HumanEval, GSM-8K, MATH), architectural features like grouped-query attention and BPE tokenization with a larger vocabulary, and notes current context window support with plans for expansion. It also references upcoming tools (Llama Guard 2, Code Shield, CyberSec Eval 2) and directs users to the model catalog, documentation, and community channels.