# Rollouts, local tools and the deadlines that matter 31 October 2026 · Evening edition Coverage: 2026-10-31T05:35:00Z to 2026-10-31T17:30:00Z ## The take Rollout boundaries matter more than model launches this evening. Apple sets a China timetable with explicit exclusions, while infrastructure users face secret rotation and service deadlines. Local tools keep broadening their capabilities, but product previews still need separating from usable, tested releases. ## Apple sets a November beta for China CONSUMER AI / ANNOUNCED 31 OCT / BETA 4 NOV - Apple says its first core Apple Intelligence features for mainland China enter the iOS developer beta on November 4, followed by iOS 27.3 in December. This is a rollout plan, not availability today. - The initial set includes live translation, Writing Tools, Photos Clean Up and Genmoji. Alibaba Qwen connects to selected Siri, Writing Tools and Visual Intelligence experiences. - Siri AI and the other next-generation features announced at WWDC 2026 are explicitly excluded. Other platforms await necessary approvals; supported iPhones and language settings also limit eligibility. [Apple announcement (Chinese)](https://www.apple.com.cn/newsroom/2026/10/apple-intelligence-features-are-coming-to-users-in-china/) · [Apple China feature overview](https://www.apple.com.cn/apple-intelligence/) · [Independent coverage / SCMP](https://www.scmp.com/tech/big-tech/article/3372831/apple-intelligence-roll-out-iphones-china-december-alibaba-backing) ## A small OpenCL addition LOCAL INFERENCE / 31 OCT, 08:07 UTC - llama.cpp prerelease b12020 adds OpenCL support for FWHT, the fast Walsh-Hadamard transform operation. This is a backend compatibility addition, not evidence of a general inference-speed increase. [b12020 prerelease](https://github.com/ggml-org/llama.cpp/releases/tag/b12020) · [Implementation / PR 31434](https://github.com/ggml-org/llama.cpp/pull/31434) ## Print preparation gets clearer warnings 3D TOOLS / 31 OCT, TIME UNVERIFIED - 3D AI Studio v6.9.2 highlights thin walls and offers strengthening while preserving colour and texture. Its repair panel explains fixes such as closed holes and reversed faces. - Export size checks now align to a tenth of a millimetre, and downloads wait for pending geometry/colour updates. These are vendor release claims, not a test of printability; the day-only timestamp does not prove a post-morning launch. [3D AI Studio changelog](https://www.3daistudio.com/Updates) ## Northflank expands secret-rotation tools UPDATE / 31 OCT, 13:34 UTC / INCIDENT OPEN - MySQL and Redis secret rotation is now available to all customers, Northflank says. Its incident remains Identified / Major outage; the tooling update is not a resolution or a newly disclosed compromise. - The provider still advises rotating all secrets. Revoke exposed credentials with the issuing provider, create replacements, update affected workloads and inspect AI-provider usage and billing. Earlier guidance also calls for disabling public access to addons. - Audit logs became available to all customers at 01:04 UTC. Separately, customer Quiltt reports rotating exposed credentials and finding no evidence of unauthorized customer-data access in its own review. That result does not establish the impact across Northflank. [Northflank incident timeline](https://status.northflank.com/) · [Customer cross-check / Quiltt](https://status.quiltt.io/incidents/01M67Q509HGBPBB9KPP2GHACKW) ## Smithery proxy users reach migration day DATED CATCH-UP / SHUTDOWN DATE 31 OCT / HOUR UNSPECIFIED - Smithery scheduled its MCP Proxy and managed identity/connection services to shut down today. Its notice includes run.tools and server.smithery.ai endpoints, API, Connect, Toolbox, accounts and keys; CLI and SDK versions are to be deprecated. - The public, non-proxied MCP registry remains, powered by Arcade. Replace Smithery proxy URLs with each server vendor's official MCP endpoint. This is an earlier-announced deadline, not an independently tested claim that every endpoint has already stopped. [Smithery transition notice](https://www.smithery.ai/future) ## KoboldCpp adds more local media and agent tools MISSED MORNING CATCH-UP / 31 OCT, 04:51 UTC - Version 1.124 integrates audio.cpp models, including YuE2 music generation and TTS options, while MusicUI gains a transcription tab. It also adds support for image models including MOSS Image, Fibo 2 Edit and Flux 3 Klein. - For agents, the release adds a custom MCP endpoint wizard and fixes an MCP loader race. Kobold Lite gains /v1/responses support and shared parallel context-pool controls. - The notes also list CPU flash-attention and legacy-model buffer-overflow fixes. These are maintainer-stated capabilities, not our benchmark results; hardware needs and individual model licences still vary. [KoboldCpp 1.124 release notes](https://github.com/LostRuins/koboldcpp/releases/tag/v1.124) ## EINSTAR expands the mobile scanning workflow MISSED CATCH-UP / RELEASED 29 OCT - EXStar Hub 2.0 and Android EXStar Mobile add a mobile workflow for Rockit and EINSTAR 2. iPhone and iPad support is still forthcoming. - The vendor says wireless IR without texture and Laser Feature capture rise from 20 to 30 FPS; IR Texture is not included. Real-Time AI Isolation needs Windows with an RTX 3060 or higher; Mac support is in development. - 3D Gaussian Splatting is for visual presentation, not a replacement for a manufacturing/measurement mesh. Each project stores one mesh, so importing a replacement overwrites it. [EINSTAR release details](https://www.einstar.com/ja/blogs/news/exstar-hub-2-0-update) · [Publication date / news index](https://www.einstar.com/blogs/news) ## A desktop metal printer, still a launch plan MISSED CATCH-UP / REVEALED 28 OCT - Studio 3e8 plans a December 1 Kickstarter launch for Mesa1, starting at $3,500. That is a planned crowdfunding price, not shipping availability. - The company describes sealed metal/scaffold cartridges depositing into a crucible inside an integrated sintering furnace, without lasers or binders. Its samples and process claims have not been independently tested here; a reveal is not proof of reliable production. [Studio 3e8 original reveal](https://www.studio3e8.com/post/studio-3e8-launch) ## DeepSeek's free web-call period has ended on its published schedule UPDATE / SCHEDULED CHANGE 31 OCT, 16:00 UTC - The October 17-announced free period for server-side Search and Fetch calls ended at 16:00 UTC today (November 1, 00:00 Beijing), according to DeepSeek's schedule. Each endpoint is listed at $0.75 per 1,000 successful calls, plus model-token charges. - Only HTTP 200 responses with nonempty results or fetched content count; failures, empty results and timeouts do not. Documentation still uses forward-looking language, and we have not verified a charged transaction. [DeepSeek Search/Fetch rules](https://api-docs.deepseek.com/guides/search/) · [Original 17 Oct notice (Chinese)](https://api-docs.deepseek.com/zh-cn/news/news261017/) ## US student offer reaches its claim deadline DEADLINE CATCH-UP / CLAIM BY 31 OCT / TIMEZONE UNSPECIFIED - Eligible students at US degree-granting colleges and graduate schools can claim four free monthly Plus billing periods through SheerID. The offer is not a UK-wide student promotion. - A valid payment method is required; it renews at $20 per month unless cancelled. Existing direct-billed monthly Plus users can qualify, but App Store/Google Play-billed subscriptions are excluded while store-billed. The official deadline gives no exact hour or timezone. [Official eligibility and terms](https://help.openai.com/en/articles/20001493-chatgpt-back-to-school-offer-for-students)