Alibaba's Qwen4-Max reasoning model with native vision and a 1M-token context window is now available on AI Gateway.
Observability changelog digest
Nov 9 - Nov 15, 2026. Useful releases, risky migrations, and noisy updates from the Observability channel.
No matching updates in this bucket.
Observability updates in 2026-w46
Daemons can now send OpenTelemetry traces to Vercel Tracing via a Unix socket provided at VERCEL_OTEL_SOCKET_PATH.
Vercel Token Pass now includes $150 of tokens per month for $30/team/month, up from the previous $80 limit.
Browser Run Quick Actions now support local development using a local Chrome instance with `npx wrangler dev`.
Set account-level spending limits and email alerts for Workers and Durable Objects to control costs.
Parallel Decisions Fast is now available on AI Gateway for free through November 23, 2026, using the model name parallel/parallel-decisions-fast.
Evaluation creation has been reworked for clearer choices, reusable starting points, and fewer surprises.
LLM judges can now send images, PDFs, audio, and video from evaluated events as attachments.
Manage AI Search instances and namespaces in any namespace using Terraform with the new namespace argument and resources.
Vercel Security Review no longer requires a linked Vercel project to scan code changes in GitHub repositories.
The build cache size limit on all Vercel build machines has been increased to 5 GB automatically at no additional charge.
Vercel Sandbox now initializes up to 4x faster outside the Washington, D.C. region by using regional caches for reused image data.
Local evaluation cache hits now count as one flag request to fix unbilled usage.
Kimi K3.1 Pro is now available in Token Pass for Pro teams, offering coding, agentic workflows, vision, tool use, and a 1M context window.
AI Search now supports D1 databases as a data source in beta, indexing each row of a read-only query as a separate document with scheduled syncs.
Vercel Sandbox now supports installing private npm packages and custom registries using team Shared Environment Variables.
Improved fingerprint rules to make them easier to write and debug.
Inkling 2 Small from Thinking Machines is now available on AI Gateway for multimodal reasoning, coding, and visual tasks.
Connect Circleback as a Data Warehouse source to sync its data into PostHog.
Google's Gemini 4 Hydrogen model is now available on AI Gateway under the name google/gemini-4-hydrogen.
Added support for exporting D1 databases containing virtual tables like FTS5 and R*Tree by setting compatibility options.
Added noise cancellation to web apps built on RealtimeKit using the new @cloudflare/realtimekit-noise-cancellation package.
Data Warehouse setup now recommends incremental fields prioritizing update timestamps.
Expanded sources to sync data from Stripe, HubSpot, Postgres, Salesforce, and over 1400 other sources into PostHog or your own warehouse.
Google's EmbeddingGemma 2 multimodal embedding model is now available on Workers AI.
Postgres change data capture (CDC) is now generally available for eligible tables without a feature-flag gate.
Unneeded CDC history tables are now deleted from the Data Warehouse schema configuration instead of being retained indefinitely.
Run `wizard support` to enable PostHog Support, configure its web widget, and check SDK initialization.
Python 3.15 is now available as a Vercel Managed Image for Vercel Sandbox.
Data Warehouse person-property sync runs now show their progress when interrupted.
RealtimeKit auth tokens now support custom validity windows and individual token revocation.
Multiple data warehouse sources including object storage, databases, and business tools have reached general availability.
Kitesurf, Cloudflare's browser engine for AI agents, is now open source on GitHub under the Apache 2.0 license.
Added the Inkling 2 Small multimodal Mixture-of-Experts model to Workers AI, featuring 594B total parameters, vision understanding, reasoning, function calling, and a 1M token context window.
Introduced a new Chrome Toolbar extension as an alternative to the Toolbar.
Expanded Data Warehouse sources with a large batch of new tables across marketing, commerce, engineering, and operations tools.
Redesigned the product sidebar into a five-item rail and updated the Home view with pinned metrics and daily product changes.
Wrangler 4.160.0 and Cloudflare CLI 1.0.0-beta.29 now support custom API request headers via the CLOUDFLARE_API_HEADERS environment variable.
Google's Gemini Decision 1 multimodal model is now available on AI Gateway for routing, policy checking, tool filtering, and ranking without output tokens.
Serve Worker static assets from a URL path prefix like /blog/ without changing your build output by configuring assets.base_path or worker.assets.basePath.
Workers Paid plan customers can now create up to 100 Hyperdrive configurations per account, increased from 25.
Reduced round trips and intermediate connections when connecting to relocated Durable Objects.
Added a collaborative whiteboard plugin for RealtimeKit meetings hosted and managed by Cloudflare.
Speed Insights and Web Analytics now fully support custom environments, attributing data previously sent to Preview to the correct custom environment.
Introduces an official Rust SDK for Flagship, enabling async server-side flag evaluation via OpenFeature.
Workers Builds and Pages now list GitHub repositories based on your own GitHub account and accessible organizations.
MiniMax H3 Image and MiniMax H3 Image Highspeed are now available on AI Gateway with support for text-to-image generation, image editing, custom dimensions, and 2K/4K outputs.
Generating MP4 downloads from Cloudflare Stream is now up to 40% faster without requiring any implementation changes.
Google's Gemini 4 Carbon model is now available on AI Gateway with a 50% discount through November 30, 2026.
Set a concurrency limit on individual Workflows to cap concurrent instances, protecting downstream systems and distributing load.
Snapshots in Vercel Sandbox are up to 10× faster, with median creation time under 0.4 s (down from 4.4 s).
Build disk space doubled to 64 GB on Basic and Standard machines.
Workflow SDK v5.2.0+ now sends event and stream writes to Vercel Workflows over a WebSocket by default, replacing per‑write HTTP requests.
Muse Ember, a lightweight reasoning model from Meta, is now available on AI Gateway with support for text, image, video and a 1 M‑token context window.
Added support for running Nanocode coding agents in Cloudflare Durable Objects with persistent session history.
Added support for running Exo agents on Cloudflare via a new adapter, using Durable Objects and Sandboxes.
Durable Objects now preferentially restart on the same machine they last ran on, reducing cold start overhead.
Claude Fable 5.2 is now available on AI Gateway, offering improved coding, reasoning, vision, faster output, and long‑running task orchestration.
Container Registry now supports compressed image layers up to 2 GB (up from 500 MB) with a 15 GB total image limit.
Introduced an experimental Agents SDK example to run the Codex agent loop in a Cloudflare Durable Object, persisting conversations, files, and model/tool results across restarts.
Integrates Firebase Genkit flows with PostHog AI observability for tracing model, tool, and embedding calls.
GLM 5.3 Flash is now available on Token Pass alongside Kimi K3, GLM 5.3, and DeepSeek V4.1 Flash.