Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-20 · source-backed
Cloudflare announced Workers AI now supports large-scale model inference, starting with Kimi K2.5 (256K context, vision, multi-turn tool calling). Internal testing: switching a 7-billion-token/day security code review agent from a proprietary mid-tier model to Kimi K2.5 achieved a 77% cost reduction. New prefix caching with session affinity headers and redesigned async APIs target the cold-token-cost bottleneck that kills agent economics at volume. Cloudflare just shifted from CDN to full-stack AI inference layer.
Each link below shares sources, entities, or timing with this story.
Cloudflare added Moonshot AI's Kimi K2.5 to Workers AI on March 19, making it the first frontier-scale open-source model available on edge compute with a full 256K context window, multi-turn tool calling, vision inputs, and structured outputs (Cloudflare Blog). Cloudflare repo...
Cloudflare just dropped a dozen infrastructure products for agent builders in a single week, and the one that caught my eye isn't the flashiest. Unweight achieves 15-22% lossless LLM compression by decompressing weights directly into on-chip shared memory, bypassing slow main...
Ryan Dahl announced celld on August 5. It's a daemon built from V8, S3, SQLite, LTX and Tokio that runs the exact Cloudflare Workers and Durable Objects JavaScript APIs and configuration on hardware you own. The architecture is the interesting part, not the API compatibility....
This is the most immediately useful pattern I've seen this month. Cloudflare shipped Code Mode MCP, and the numbers are hard to argue with. Their API has 2,500+ endpoints. Exposing each one as a separate MCP tool would consume 1.17 million tokens of context just for tool defin...
Cloudflare released @cloudflare/computer on August 3 as day two of Agents Week, and the motivation in the engineering post is the part that should change your architecture thinking. Not speed. Capacity. (Cloudflare Blog) Their stated position: across every hyperscaler, there i...
Cloudflare opened a waitlist for its Monetization Gateway, and it's the most interesting new distribution primitive I've seen this quarter. The pitch: charge for any web page, dataset, API, or MCP tool sitting behind Cloudflare, with charges settling peer-to-peer in stablecoin...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.