Launches and major updates that matter to people building with AI. Every item is dated, links to its primary source and gets a short verdict: is it worth trying, and for whom?
Updated Oct 22, 2026 · 24 releases tracked · RSS
Security-focused patch: upgrades the bundled Bun runtime to 1.4.4, which carries fixes for Brotli, TLS, HTTP/1.1, HTTP/2, HTTP/3, WebSocket and fetch, and refreshes root certificates.
Verdict: Every OpenCode user: a quick update for the runtime security fixes.
Bug-fix and reliability release, following v2.1.307 (Oct 21) which added /claude-api api-version, a safety_event OpenTelemetry event, and multiple MCP reconnect fixes.
Verdict: Claude Code users with MCP-heavy setups: update for the MCP reliability fixes.
Ask Lovable to create a Supabase project (you pick org and region) for an app with no backend; it connects it automatically. Same release added grouping workspaces into organizations.
Verdict: Lovable builders who want a real Postgres backend they own.
Interactive V3 sessions now ask you to trust a folder before loading its hooks, agents, MCP servers, steering and workflows. Also workflow plan step trees and Mermaid subgraphs.
Verdict: Anyone running Kiro CLI in cloned or third-party repos (security win).
Mention or assign a Linear issue to Amp and it works the issue in a thread.
Verdict: Teams that plan in Linear and want issue-to-PR delegation.
Google's frontier model is available via the Gemini API/AI Studio, in Antigravity for AI Ultra subscribers, and in the Gemini app. Introductory API price $2 input / $10 output per 1M tokens.
Verdict: Builders doing large refactors or long agentic tasks: try it before the Dec 1 price increase.
Anthropic's fastest small model, built for high-volume work like summarization, subagents and browser use. $0.50 / $2.50 per 1M tokens (prompts ≤100K).
Verdict: Builders running subagents or high-volume tasks on a budget.
See and reply to local agents running on your computer from the Cursor iOS app; on by default except Enterprise.
Verdict: Builders who kick off long local agent runs and step away.
OpenAI's latest GPT-6 family model. API: $2 input / $10 output per 1M tokens (standard, short context).
Verdict: Builders who want a strong mid-priced default model.
Clear upgrade over Sonnet 5: 30%+ faster and up to 30% cheaper for most work. API $2 / $10 per 1M tokens.
Verdict: Default model choice for most coding agents and AI features.
Leads in agentic coding; performs at Fable 5.1 level on most work and costs ~40% less to run than Opus 5. $4 / $20 per 1M tokens, cache reads $0.20.
Verdict: Long-running coding agents and big migrations.
OpenAI expanded the GPT-6 family with Sol and the low-cost Luna (API $0.10 / $0.50 per 1M tokens).
Verdict: High-volume, cost-sensitive features (Luna).
Prompt mode (-p and piped stdin) now honors managed Auto model selection, subagents respect custom and MCP tool approvals without duplicate prompts, plugin marketplaces accept git-subdir sources, and replies cut off by provider overload now retry with backoff.
Verdict: Copilot CLI users scripting it with -p or leaning on subagents and MCP tools.
Adds GPT-6 Astra and Claude models on Bedrock, smaller binaries on all platforms, and bug fixes.
Verdict: Zed users on Bedrock; others can just auto-update.
Stable release that enforces read-only workspace settings in untrusted folders, fixes session-resume bugs (duplicate tool turns, lost history on quick exit), aligns OAuth callback validation with RFC 9207, and adds custom OTLP headers for telemetry.
Verdict: Gemini CLI users who resume sessions or open untrusted repos; others can update at leisure.
Slack MCP integration lets v0 find relevant team messages and post updates to Slack when asked.
Verdict: Teams using v0 collaboratively; solo builders can skip.
GitHub deprecated several models across Copilot Chat, inline edits, ask/agent modes and code completions as of Oct 19, 2026.
Verdict: Copilot users pinned to specific models: check your model picker.
Neon now has realtime (built by the team behind Electric): turn any SQL query into a live query.
Verdict: Neon users who need live updates without adding another service.
Cloud agents can build interactive canvases and share them by link; Environments API in beta; environments up to 300 repos and 5x faster loads for large ones.
Verdict: Teams using Cursor Cloud Agents for reports and demos.
Origin (Cursor's code hosting) now integrates with 20+ apps for deploys, CI, review, security and monitoring (Cloudflare, Linear, Sentry, Socket) and opens to public submissions.
Verdict: Teams hosting code on Origin; dev-tool makers who want distribution.
New self-serve customers buy Cursor via The X Subscription: Plus $8, Premium $30, Super $100, Ultra $200 per month. Third-party models in Cursor are billed separately at API list price on Premium+. Legacy Cursor plans continue; switching is optional and permanent.
Verdict: New Cursor users who mostly use Cursor/Grok models. Heavy Claude/GPT users: compare costs before switching from a legacy plan.
GPT-6 is rolling out globally in ChatGPT with Intelligent UI.
Verdict: ChatGPT/Codex users: little action needed beyond trying it.
Anthropic's most advanced models for coding and knowledge work. Fable 5.1 API $10 / $50 per 1M tokens.
Verdict: Hardest problems only; Opus 5.5 is cheaper for most builders.
Devin Desktop v3.9.19 removes the legacy Cascade agent; Devin Local is now the only local agent.
Verdict: Former Windsurf users still relying on Cascade workflows.
No releases in this filter yet.
Get the daily Release Radar
Free. Double opt-in, unsubscribe anytime. See our privacy policy.