Model intelligence.
Follow the changes. Understand what comes next.
Why AI Router ZDR Labels Are Not a Guarantee: Inside OpenRouter's MiniMax Warning and a 4-Step Hardening Guide
Following Dragos's November 19, 2026 study of two million canary-seeded prompts, OpenRouter emailed affected customers on November 20–21 warning that prompts and outputs sent to MiniMax models from July to October 2026 may have been read by a third party—whether served directly or forwarded via AtlasCloud, Novita, or StreamLake—and removed ZDR labels from affected endpoints pending investigation.
Latest stories126
Newest publishedAmazon Bedrock AgentCore Adds Monthly Free Web Search Queries and Gateway domainOverride Support
In its November 2026 release notes, Amazon Bedrock AgentCore introduced monthly free web search queries (stated in the release notes as the first 5,000 queries per month per AWS account, excluding AWS GovCloud US) and a domainOverride setting for gateways using JWT inbound authorization.
Claude Code v2.1.326 Released: Code Mode with Agent Teams, Updated Auto Mode Defaults & Sandbox Fixes
Anthropic updated the npm latest tag for Claude Code CLI to v2.1.326 (dated November 16, 2026 in the official changelog and published as a non-prerelease GitHub Release). The release enables turning on Code Mode alongside agent teams, updates default auto mode behavior for HIPAA organizations and unloaded policies, and ships 59 improvements and 233 bug fixes.
Claude Code Desktop on macOS Adds iPhone Duo Simulator Testing (Public Beta) Alongside Official Xcode Plugin
Anthropic's documentation details the iOS Simulator pane in public beta for Claude Code Desktop on macOS, including iPhone Duo foldable posture testing on Claude Desktop v4.29.0+ and Xcode 27.1+, while the claude-plugins-official manifest lists an Apple Inc. Xcode plugin.
Notion Adds Full-Page Meeting Notes REST API and MCP Tools with Calendar Event Linking
On November 16, 2026, Notion announced REST API and Notion MCP support for full-page meeting notes. Developers can create full-page meeting notes via POST /v1/meeting_notes, link existing notes to calendar events via PATCH /v1/meeting_notes/:meeting_note_id, inspect new is_full_page and event_id fields in block responses, and use three new Notion MCP tools.
GitHub Adds strictPluginOnlyCustomization Managed Setting to Control Copilot Customization Sources
On November 16, 2026, GitHub announced strictPluginOnlyCustomization, a new enterprise-managed setting on GitHub Enterprise Cloud that requires selected Copilot customization categories—skills, custom agents, hooks, MCP servers, and executable Copilot extensions—to come from plugins or managed settings rather than standalone user or project configs. Admins can pair it with strictKnownMarketplaces to restrict eligible marketplace sources.
Google Antigravity App v2.33.0 & CLI v1.5.9: SSH to 64-bit Linux Hosts and GitHub PR Status
Google Antigravity's official changelog now marks App v2.33.0 (dated November 17, 2026; no official clock time) as Latest, adding SSH connections to remote 64-bit Linux machines plus live GitHub pull request status pills, hover cards, and an Overview Pull Requests section. The same changelog also lists App v2.32.2 (dated November 16, 2026), which brings multi-turn /btw side-question tabs and a diagnostic log truncation fix. The CLI Latest remains v1.5.9 (November 14, 2026). The earlier App v2.32.1 (November 13, 2026) added Gemini 4 Hydrogen (low, medium, or high thinking levels) to the model selector when signing in with Google Cloud application default credentials; CLI v1.5.9 introduced the bundled-Python run_workflow subagent tool, pluginControls for non-Gemini Enterprise organizations, and an HTTP MCP redirect header security fix.
Devin Desktop v3.13.4 & CLI v3000.13.4 Release Notes: Multi-Tab Command Center & Shared Background Sessions
Cognition's stable changelogs list Devin Desktop v3.13.4 (November 17, 2026) and Devin CLI v3000.13.4 (November 16, 2026). Both clients now run local sessions in a shared background process so closing or reopening either client keeps sessions alive across multiple windows, while Desktop adds Agent Command Center tabs and local file previews and CLI adds devin daemon commands and .kiro/skills discovery.
Tencent Hy4 flash vs. Hy4 Selection & Pricing Guide: Workload Routing, 50% Discount Deadlines & Preview Retirement
On November 16, 2026, Tencent launched Hy4 flash for high-speed, cost-efficient everyday workflows alongside its flagship Hy4 model. This guide compares official workload recommendations, TokenHub RMB catalog rates (¥0.50/¥2.00 for Hy4 flash vs. ¥6.00/¥18.00 for Hy4 per 1M tokens), separate 50% discount deadlines (Nov 30 vs. Dec 31 Beijing Time), research-page USD reference rates, and the December 7, 2026 retirement of hy4-preview.
Gemini 4 Selection & Pricing Guide: Choosing Between Hydrogen, Carbon, and Argon
With gemini-4-hydrogen (Nov 12, 2026), gemini-4-carbon (GA Nov 10, 2026), and gemini-4-argon (Oct 2026) live on the Gemini Developer API, choosing the right model comes down to Free tier eligibility and the December 1 rate change. All three share a 1,048,576-token context window (65,536 default max output; up to 1,048,576 with long decode). Hydrogen stays at $1.50/$7.50 per 1M tokens with Free tier access, whereas Carbon and Argon exclude the Free tier and rise from $2/$10 through November 30 to $4/$20 starting December 1, 2026.
How to Use the github-security Extension in the GitHub Copilot App (Public Preview)
GitHub's documentation details the github-security plugin for the GitHub Copilot app, currently in public preview. Once installed via CLI or the app sidebar, the Security canvas lets developers inspect code scanning alerts, run local agent-assisted fixes, and trigger pre-PR local security reviews—while excluding Dependabot, secret scanning, and GitHub Enterprise Server.
DeepSeek App APIs Add Web Search, File Parsing, and Speech: Integration & Pricing Guide
On November 13, 2026, DeepSeek released its full suite of application APIs. Enabling built-in flags on deepseek-flash across the Messages, Responses, and Chat Completions protocols unlocks server-handled web search, deep thinking search, file parsing, URL reading, and basic content safety checks—while Search, Fetch, OCR, ASR, and TTS are also exposed as standalone endpoints alongside an open-source chat-demo client. Here is how server orchestration splits from standalone tool chaining, which parameters get overridden, and how combined token-plus-tool billing works.
Troubleshooting OpenAI Responses API: Using input_transformations to Catch Dropped Reasoning Items
Released on November 12, 2026, OpenAI Python SDK v3.50.0 (alongside Node SDK v7.56.0) adds the input_transformations array to Responses API response objects. This field exposes which encrypted reasoning items were withheld before model execution and why—covering prefix_mismatch, turn_boundary, model_mismatch, organization_mismatch, and end_user_mismatch. This guide explains how silent reasoning drops happen during multi-turn history edits, how to trace prefix_mismatch using item IDs, and why Prompt Cache hits do not guarantee reasoning reuse.