This release adds several useful features for developers (SQLite limits, diagnostics channels, permission audit flag) and brings performance improvements and numerous bug fixes.
Preview: Fixes 155 issues (addressing 642 👍). New native REPL, –compile –target=browser for self-contained HTML, TC39 standard ES decorators, Windows ARM64 support, barrel import optimization, and a faster e
New REPL rewritten in Zig; instant start, full terminal UI, clipboard copy (.copy), top‑level await, ESM/require support, syntax highlighting, Emacs keybindings, persistent history (~/.bun_repl_history), tab completion, multi‑line input, special variables (_ and _error), proper semantics (hoisted const/let, dynamic imports, object literal detection).
Self‑contained HTML build: bun build --compile --target=browser inlines JS/CSS/assets into a single .html. CLI: bun build --compile --target=browser ./index.html; API via Bun.build({entrypoints:["./index.html"], target:"browser", compile:true}).
Full TC39 stage‑3 ES decorators now supported (accessor keyword, Symbol.metadata, ClassMethod/FieldDecoratorContext). Works for method/getter/setter, static/instance, public/private fields, class decorators, evaluation order, and legacy decorators remain functional.
Windows ARM64 support: native run on ARM64 Windows; cross‑compile via bun build --compile --target=bun-windows-arm64.
Barrel import optimization: auto‑opt for packages with "sideEffects":false; explicit via optimizeImports in Bun.build(). Improves bundler speed (e.g., lucida-react 2× faster).
Reduced bundle overhead: ESM/CJS output has ~11% fewer objects, 61% fewer getters/setters, etc.
Test retry flag: global --retry N for flaky tests; per‑test overrides; JUnit XML now reports each attempt separately.
Heap snapshot ArrayBuffer: Bun.generateHeapSnapshot("v8","arraybuffer") returns raw UTF‑8 JSON to avoid string overhead.
TLS keepalive for custom configs: mTLS and custom CA certificates now reuse connections, with LRU cache (60 entries, 30 min TTL).
Root certificate update: NSS moved from 3.117 to 3.119, removing four CommScope roots, fixing TLS failures after Cloudflare rotation.
JavaScriptCore engine upgrades:
Rope string slicing 168× faster; endsWith up to 10.5× faster; RegExp flag getters ~1.6× faster; Intl.formatToParts ~1.15× faster.
Bug fixes: RegExp.test stale captures, infinite loop with non‑greedy backreferences, incorrect backtracking, WebAssembly ref.cast/test bugs.
structuredClone & postMessage: fast path for dense arrays of primitives/strings (25× faster) and flat object arrays (1.7× faster). No regression for complex objects.
Preview: The pgAdmin Development Team is pleased to announce the release of pgAdmin 4 version 9.13. This release of pgAdmin 4 includes 15 bug fixes and new features. For more details, please see the release no
New UI/UX features – AI Controls section in Settings for managing AI‑enhanced features; New Tab wallpapers now show on container tabs.
Accessibility & privacy – Improved screen reader support for mathematical formulas in PDFs; Remote browser changes can be opted into independently of telemetry; Firefox Backup available on Windows 10 (excluding data cleared on close).
Internationalization – Added Traditional Chinese and Vietnamese translations.
Bug fixes – Language pack disabling issue after major update, correct image insertion into Adobe Illustrator on Windows, various security fixes.
Web Platform enhancements
about:blank now Web‑compatible (synchronous first navigation).
Service worker support for WebGPU in all worker contexts.
Iterator APIs: Iterator.zip() and Iterator.zipKeyed().
Trusted Types API, Sanitizer API (element.setHTML, document.parseHTML).
New properties: location.ancestorOrigins; NavigationPrecommitController.addHandler().
CSS updates: position-try-order (Anchor Positioning) and shape() function.
Community – 18 first‑time contributors joined the project in this release.
📁 Tooling
Git and GitHub
📰 GPT-5.4 is generally available in GitHub Copilot
Preview: GPT-5.4, OpenAI’s latest agentic coding model, is now rolling out in GitHub Copilot. In our early testing of real-world, agentic, and software development capabilities, GPT-5.4 consistently hits new r
Edit/Ask Modes: Edit mode hidden by default; ask mode now a custom agent. Custom agents can be created via chat.
AskQuestions Tool: Moved to core, supports steering mid‑carousel and keyboard navigation (Alt+N/P).
Session Management:
Fork chat sessions (/fork or “Fork Conversation” from checkpoint).
Inline chat now queues into existing session after file changes.
Session memory preserves plans across turns; context compaction available manually and automatically.
Browser Automation: New tools for integrated browser (navigation, read/screenshot, click/hover/type, runPlaywrightCode) with private in‑memory sessions by default.
Usages & Rename Tools: Updated usages tool and added rename tool for precise refactoring.
Explore Subagent: Plan agent delegates research to Explore (fast models like Claude Haiku 4.5 or Gemini 3 Flash), can be overridden via chat.exploreAgent.defaultModel.
Miscellaneous:
Kitty graphics protocol support in terminal.
Prevent auto‑suspend during chat requests.
New settings for agent plugins, browser tools, and session features.
New model GPT‑5.4 “Thinking” – combines GPT‑5.3‑Codex coding power with enhanced reasoning, tool use, spreadsheet/presentation support, upfront planning of thoughts, better deep web research and context retention.
Codex app releases – Windows (March 4) and macOS (Feb 2) desktop apps for running multiple Codex agents in parallel with isolated worktrees, diff review, pull‑request export; available on plans that include Codex (Free/Go trial, Plus/Pro 2× limits).
GPT‑5.3 Instant – updated March 3 to improve accuracy, web‑search results, tone, relevance and conversational flow.
Web & Android UI updates – image prompt editing, faster sharing, export of Code Block visuals, smoother scrolling, improved dictation interface (Feb 27); file upload limit raised to 20 files, expanded formats, better “Select All” behavior; Android quick‑tools menu added.
Projects & Sources – Feb 25 adds ability to add sources from apps, chats or ad‑hoc text to build a living knowledge base.
📁 Cloud
AWS
📰 Multi-party approval now supports approval team baselining
Preview: Multi-party approval (MPA) now supports MPA administrators running test approvals to confirm that their approval team is set up correctly and that approvers are active and reachable. With this new cap
Preview: AlloyDB for PostgreSQL Feature AlloyDB enhanced backups are generally available (GA). You can now select the Enhanced tier during cluster creation, manage your project-level backups with tiered tabs,
Preview: This week’s newsletter describes a standard for verifying VTXOs across different Ark implementations and links to a draft BIP for expanding the miner-usable nonce space in the block header’s nVersion
