|
The Daily Reconcile.
All the cloud native news that’s fit to reconcile
| Friday, October 9, 2026 |
Two pages |
Front page · tap to read the PDF
New models
Cloudflare has open-sourced a decision model. Clef (27B, with a vision encoder) and Clef-flash (9B) take a state plus a schema of typed questions and return a probability for every allowed option in one forward pass: no free text, no output parsing. Clef carries a 64K context against 32K for TypeSafe's Jev, whose API it copies, and median latency is 38.8 ms for Clef-flash against 524.1 ms for Jev. The weights ship under Apache 2.0 on Hugging Face. Jev lists at $0.042 per million input tokens, OpenAI's Decisions API at $0.10.
Decision latency, in milliseconds | Clef-flash · median | | 39 | | Clef-flash · p95 | | 122 | | Clef 27B · median | | 209 | | Jev · median | | 524 |
Bars to one scale, highest value 524 |
Also in today’s two pages
| → | A shared Dragonfly cache cut a 268 MB model pull from 11 seconds to 3 across two vCluster tenants; origin bytes were never metered. | | → | Reflection's Beam: 501B parameters, 23B active, pretrained on 23.8 trillion tokens in under four weeks on 6,144 GB300s. | | → | Harbor 2.15.4 puts back the two defaults 2.15.3 broke three days earlier. | | → | Dell advisory DSA-2026-448: two CVSS 10.0 flaws in Container Storage Modules, one giving root on cluster nodes. Fixed in 1.18.0. | | → | Krateo PlatformOps: frontend 1.7.0 and installer 0.3.407. |
| | kubectl tip kubectl events --types=Warning |
|
Want it on paper?
The PDF is laid out for exactly this: A4, two pages, printed on one sheet. It is how this paper is read every morning.
A4 · double-sided · flip on the long edge
|
Researched, written and laid out by Claude Code. Every item links to its source; always check the original.
Made by Diego Braga, who decides what it covers.
Sources today: Cloudflare, InfoQ, Cloud Native Now, Harbor, Reflection AI, CNCF, Dell, wolfSSL, OpenBao, GitHub and kube.events
You get this because you asked for it, and for nothing else. We keep your address and your paper preference, and no other data.
Privacy ·
Source on GitHub
|