Models advance. The systems around them must follow.

Kimi K3.1 Pro goes live; hosted access precedes weights

2026-10-24 · New

Kimi released K3.1 Pro and Highspeed with a one-million-token context window across the web, mobile apps, Kimi Work, Kimi Code and API. Full model weights are promised by November 4; enterprise administrators must enable access, which is off by default at launch.

Teams can now test long-document and cross-file tasks, while self-hosting still awaits weights. Performance and speed figures are vendor-reported; assess task completion, context use and total cost on your own workloads. API availability and weight availability are separate milestones.

Kimi: K3.1 Pro release and availability

Kimi Code release notes

Kimi official blog dates

Agent Skills moves to AAIF for neutral cross-tool governance

2026-10-22 · Recent

Anthropic announced the donation of the Agent Skills name, logo, GitHub organization and domain to AAIF under the Linux Foundation. It says more than 70 agent platforms have integrated the standard; using or contributing to it does not require foundation membership.

Neutral governance can support continued reuse of internal skills across tools. Versioned releases, a formal proposal process and security guidance remain a proposed roadmap; transferring the standard does not certify the safety of individual skills.

Anthropic: Agent Skills donation to AAIF

Codex 0.164.0 brings reusable prompts and enterprise access into terminal workflows

2026-10-23 · New

The stable release adds ten numbered /presets prompt slots, inline model and permission controls, and enterprise MCP authorization improvements. Release notes also describe credential masking for supported remote Bash/Zsh sessions and state fixes after failed conversation compaction.

The value is less repeated setup and easier recovery during long tasks. Credential masking covers supported paths, not a guarantee that every log is free of sensitive data. Teams should verify their own tool permissions and recovery workflows.

Codex 0.164.0 release notes

Kimi · Moonshot AI

K3.1 Pro / Highspeed · 2026-10-24 · stable

Hosted release is available with a million-token context; enterprise access is off by default and can be enabled by administrators. Weights are still promised by November 4.

Kimi: K3.1 Pro release and availability

Kimi official blog dates

Checked 2026-10-24T02:11:12.725491+00:00

Codex · OpenAI

0.164.0 · 2026-10-23 · stable

Stable: Prompt presets, inline controls and enterprise MCP authorization improvements. See AI Companies.

Codex 0.164.0 release notes

Checked 2026-10-24T02:11:12.725491+00:00

Claude Code · Anthropic

2.1.310 · 2026-10-24 · stable

Stable: Extends Monitor availability and fixes symlink permission prompts, a Windows hook-approval path and unintended plan approval from VS Code feedback.

Claude Code 2.1.310 release notes

Checked 2026-10-24T02:11:12.725491+00:00

JCET reports 54% computing-electronics growth; capacity still needs delivery

2026-10-22 · Recent

JCET reports third-quarter revenue of about CNY11.425 billion, up 13.53% year on year. Computing electronics represented 30.9% of nine-month revenue and grew 54%. Its Lingang advanced-packaging project is progressing, while related expansion financing still requires approvals.

These operating figures help track compute demand reaching packaging, but computing-electronics revenue is not all AI revenue. The quarterly report is unaudited; the project and financing plans began in June and September. This item covers the new operating disclosure, not planned capacity as already operational.

JCET third-quarter 2026 report

Cosmos expands five data blueprints for hard-to-capture robot scenarios

2026-10-21 · Recent

Cosmos Data Factory offers five blueprints for driving datasets, image attributes, event videos, defect images and video augmentation. NVIDIA reports that ANYbotics generated more than 1,500 synthetic clips to help industrial-inspection robots recognize liquid spills and other scene changes.

The aim is better coverage of rare events, not reliable robots simply from more videos. Generated outputs still need validation, real-data integration and task-level evaluation. Projected testing and development savings are not independently reproduced outcomes.

NVIDIA: Cosmos data and model tools

Minerva connects GPUs, networks and jobs to locate bottlenecks

2026-10-22 · Recent

NVIDIA launched Minerva, an AIOps platform connecting GPU, network, storage, scheduler and workload telemetry in a dynamic knowledge graph to correlate alerts, detect anomalies and identify likely root causes. IREN is using it; partners including IBM and Lambda plan deployments.

Beyond buying compute, operators need to measure useful work and recovery time. Minerva can recommend remediation or connect to automation, while teams can keep execution with people. Recovery and throughput benefits still need validation on the actual cluster.

NVIDIA: Minerva operations platform

OpenAI storage lessons: controlled sharing can beat more isolation

2026-10-23 · New

OpenAI describes Habitat moving from shared to dedicated containers, then back to sharing with hot-key throttling, request priorities and improved serverless partition scaling. Dedicated containers reduced failure spread but pushed capacity planning onto individual product teams.

For large AI services, data-layer operational complexity can slow product iteration. The article also discusses caching and cross-region capacity-versus-latency tradeoffs. A next-generation system based on RocksDB Cloud and Raft is still being developed, not a completed replacement.

OpenAI: Storage at one-billion-user scale, part two

China brings employment effects into AI project assessment

2026-10-23 · New

China’s State Council General Office published employment-friendly AI guidelines. They call for employment-risk assessment alongside review of market-entry projects that may cause substantial job displacement, and encourage employers to explore transition buffers, advance notice and redeployment support.

An AI deployment plan should therefore include training, job transitions and worker protections alongside procurement and efficiency targets. The document was dated October 16 and published October 23. Encouraging transition buffers does not impose one immediate, uniform duration on every employer.

State Council: Employment-friendly AI development

US proposes broader outbound-investment rules for AI and compute

2026-10-22 · Recent

The US Treasury proposed COINS implementing rules covering technologies including semiconductors, AI, quantum and high-performance computing, with biotechnology also proposed. The framework distinguishes notifiable and prohibited transactions and addresses covered persons, exceptions and reasonable pre-transaction inquiry.

Cross-border investors need to examine technical parameters, transaction structure and covered persons, rather than a blanket notion of an AI investment ban. This remains a proposal for comment; final regulations and their effective date are still to follow.

US Treasury: Proposed COINS implementing rules

Solidigm ends its IPO review while considering other funding routes

2026-10-23 · New

SK hynix announced that Solidigm will stop reviewing potential pre-IPO and IPO options, citing capital-market conditions and share-price trends. It still sees a need for additional US production facilities; investment needs are unchanged and alternative funding arrangements remain under consideration.

Memory demand and financing routes can move at different speeds. Ending a listing review does not cancel factory investment or mean a filed IPO was withdrawn. The announcement confirms neither alternative funding amounts nor commissioning dates or new available capacity.

SK hynix: Solidigm discontinues IPO review

SK hynix dated news index