AI coding tools news
November 18, 2026 · 20 stories
The day's news for developers who use AI coding tools, picked and summarised from official changelogs, the media and Hacker News. Resets and limit changes are on the reset tracker.
Claude Code updates model switching and Messages API thinking blocks
Claude Code updated conversation model switching for Opus 5, Opus 5.5, Fable models, and Sonnet 5.5. It also updated how the Messages API handles thinking blocks to preserve thinking and protect against distillation.
GitHub Copilot adds eight new MCP connectors in public preview
GitHub Copilot now supports connecting eight additional services, including Supabase and Box, using GitHub-verified Model Context Protocol (MCP) servers in public preview.
Google launches Gemini 4 Neon in Gemini, Antigravity, and APIs
Google launched Gemini 4 Neon priced at $3 per 1 million input tokens and $15 per 1 million output tokens. The model is now available in Gemini, AI Studio, and Antigravity.
Claude Code adds project organization and updates Messages API thinking
Claude Code adds task organization through projects in Claude Cowork and updates Messages API thinking blocks handling. Changes also include model configuration adjustments and updates regarding model switches.
Snapchat lets developers build AR lenses with Claude Code and Cursor
Snapchat announced that lens creators can describe the desired lens in plain language and an AI assistant will handle the process from start to finish. The AI assistants mentioned are Claude Code and Cursor. The feature enables developers to build AR lenses using these coding tools.
GitHub Copilot adds team customizations feature
GitHub Copilot now lets organizations share plugin setups across team members. The team customizations feature is generally available for organization teams. Users can apply shared plugins without individual configuration.
Qwen4-Max preview and release dates, pricing vs Claude Opus 5.5
Qwen4-Max became available in preview to Token Plan subscribers on November 11 and was generally released on November 16. The article compares its cost to Claude Opus 5.5, noting the cheaper model incurs costs elsewhere.
CLI: Organization MCP Sign-In and Custom Login Portals
This release adds organization sign‑in for MCP servers in V3 sessions and a configurable portal for kiro‑cli login. One sign‑in through your organization’s identity provider now covers every server set to use it, and administrators can choose where the browser sign‑in starts.
Codex sandbox enforces filesystem and network limits across OSes
Codex runs commands on your machine while enforcing filesystem and network limits you control across macOS, Linux, and Windows. The limits are configurable by the user. No specific numeric limits are provided in the text.
Google Loop autonomous agent monitors live data and opens PRs
Google Labs announced Stitch Loop, an autonomous product agent that monitors live data and automatically opens pull requests for security, latency and accessibility issues. The announcement was made in an early‑look article on TestingCatalog. No pricing or usage limits were provided.
Mistral launches Vibe desktop app for macOS and Windows
Mistral releases Vibe as a desktop app for macOS and Windows. It runs coding agents locally or in the cloud. Free Vibe Pro is available in Europe.
Claude Opus 4.8 and 4.7 launch with 1M token context and pricing details
Claude Opus 4.8 (claude-opus-4-8) supports a 1 M token context window and 128 k max output tokens on Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Claude Opus 4.7 is also launched with the same $5 / $25 USD per MTok pricing as Opus 4.6 and includes an updated tokenizer and API breaking changes. Memory for Claude Managed Agents is now in beta under the managed‑agents‑2026‑04‑01 header. The Messages API is now in research preview on Amazon Bedrock in us-east-1, with a new /anthropic/v1/messages endpoint.
Google Gemini 4 Neon pricing added to Cloud SKU and Antigravity CLI
Google added Neon pricing to its Cloud SKU and to the Antigravity CLI.
Google Gemini 4 Neon pricing added to Cloud SKU and Antigravity CLI
Gemini 4 Neon appears in Google Cloud’s SKU catalog with prices of $3 and $15 per million tokens. It is also listed in the Antigravity CLI as a Pro orchestrator model.
OpenAI API Pricing Update – New GPT‑6.2 Astra/Sol Rates and Ultrafast Multipliers
The pricing page lists per‑1M‑token rates for GPT‑6.2 Astra ($8 prompt, $20 completion) and GPT‑6.2 Sol ($2 prompt, $5 completion) under the Standard tier. Ultrafast mode costs three times the Standard rate until Feb 28 2027 and six times thereafter. Prompt‑caching rates are lower (e.g., Astra $4 prompt, $10 completion). Background request discounts give 30% off Standard for 2‑hour windows and 40% off for 8‑hour windows. Regional data‑residency and FedRAMP endpoints add a 10% uplift for models released after Mar 5 2026. Additional multimodal model prices and legal‑model rates are also listed.
Microsoft launches MAI Code 2 and MAI Code 2 Flash for Copilot
Microsoft launches MAI Code 2 and MAI Code 2 Flash models for GitHub Copilot.
Codex 0.170.0 release adds WebRTC voice, auth header refresh, MCP support
The 0.170.0 release adds WebRTC negotiation for realtime voice calls in the TUI. It also adds automatically refreshed authentication headers for OpenTelemetry HTTP exporters and experimental support for using identically named MCP servers across environments. Bug fixes include fallback of oversized WebSocket requests to HTTP, MCP OAuth credential race fixes, improved Cloud conversation reconnection, and various Windows sandbox and permission fixes. Documentation was updated for sandbox usage.
Kiro Enterprise adds profile migration to Enterprise Management
Enterprise administrators can now move existing Kiro profiles to Enterprise Management. The preview is limited to US East (N. Virginia) (us-east-1). Profiles can use IAM Identity Center or an external identity provider.
Kiro Enterprise adds local sandbox controls for IDE and CLI
Enterprise administrators can now require sandboxing for commands and local stdio MCP servers in Kiro IDE and Kiro CLI. They can set default file and network permissions, and developers cannot disable a required sandbox via client settings. Organization‑set deny rules override any allow rules, and credential stores remain closed unless an explicit file allow rule opens them.
Claude Code release notes: model, effort, and thinking settings changes
Claude Code release notes indicate changes to the model, effort, and thinking settings. They also show which Claude Fable models are available on a user's plan and provide information on how up‑to‑date the training data is.