What changed in the major AI coding CLIs?

OpenAI updated Codex with default plugins, npm marketplace integration, MCP tool search for efficiency, system proxy support, Amazon Bedrock support, and maximum reasoning effort [4174]. The Codex Tag demo has been rewritten to be lighter (core under 10k LOC) and perform better, delegating long running tasks to pidotdev [4175]. OpenAI Codex team turned code review into an AI that checks intention behind code, catching bugs automatically [4173]. Anthropic released an official skill called Frontend Design for Claude Code that forces aesthetic direction before coding, generating production-ready HTML/CSS/JS/React/Vue [4169]. Claude Code artifacts gain ability to call MCP connectors, enabling dynamic dashboards and apps that fetch information and take actions per viewer [4168]. Claude Code and Codex can now directly call Seedance 2.0 for video generation, with API costs significantly lower than competitors like Higgsfield [4170]. Claude AI introduces Claude for Teachers, offering free premium access to verified K-12 educators in the US with teaching skills and curricula mapping [4167]. Gladstone Institutes prize awarded to Trialign, a clinical trial matcher for cancer patients built with Claude [4172]. Google Gemini CLI received multiple fixes: strip thoughts from scrubbed history turns and resolve thought leakage [4166], preserve escape sequences in string literals for modern models [4165], make ~/.gitconfig read-only in the macOS sandbox [4164], egress cloud run service skeleton [4163], resolve symbolic link directory escape in memory import processor [4162], implement Cloud Run webhook ingestion service [4161], resolve defensive path resolution for at-reference files and fix macOS tests [4160], enforce case-insensitive sensitive path blocklist and vscode hitl [4159], and fix no_proxy test [4158]. Claude Code desktop app received many fixes and features: home page command palette search can now find and open sessions [4157], fixed custom agent selector visibility when custom agents are available [4156], fixed startup readiness so WSL server loading is included before the desktop app reports ready [4155], improved dangerous-command detection including more forced rm forms and clearer rejection reasons [4154], fixed very large markdown tables stalling rendering (tables over 200 rows show first 200 with notice) [4153], fixed truncated stream-json/JSON output when piping large responses from claude -p [4152], fixed replies typed to a background agent being lost when delivery fails [4151], fixed fast mode staying off after switching back to a model that supports it [4150], added mouse-click support for multi-select menus and "Other" input rows in fullscreen mode [4149], added vimInsertModeRemaps setting [4148], added screen reader mode [4147], fixed plan approvals without edits being labeled "(edited by user)" [4146], fixed hook callback timeout misreported as user rejection [4145], fixed session crash when tool result renderer returned bigint or plain text [4144], fixed paste markers leaking into external editors [4143], fixed rendered text fragment leaking into crash telemetry [4142], fixed ultracode keyword opt-in firing on non-human-originated input [4141], added startup warning for Write/NotebookEdit/Glob permission rules [4140], added live elapsed-time counter to collapsed tool summary line [4139], fixed edits that leave input as "?" being silently swallowed [4138], fixed file upload validation for DOS device suffixes and multiple hard links [4137], fixed Claude Code on Vertex and Bedrock attempting default Opus model at startup [4136], fixed auto mode overriding PreToolUse hook's ask decision for unsandboxed Bash [4135], fixed permission previews not neutralizing bidirectional-override characters [4134], fixed shell mode not executing commands containing file paths while autocomplete popup was open [4133], fixed SIGTERM during Bash tool orphaning process tree [4132], fixed continue:false hook's halt being dropped [4131], fixed plan mode auto-running file-modifying Bash commands without permission prompt [4130], added session-wide limit on WebSearch tool calls (default 200) [4129].[46]

OpenAI updated Codex with default plugins, npm marketplace integration, MCP tool search, system proxy support, Amazon Bedrock support, and maximum reasoning effort [4174]. The Codex Tag demo was rewritten to be lighter and delegate long tasks to pidotdev [4175]. OpenAI Codex team turned code review into an AI that checks intention behind code [4173]. Anthropic released Frontend Design skill for Claude Code [4169]. Claude Code artifacts gained ability to call MCP connectors [4168]. Claude Code and Codex can now call Seedance 2.0 for video generation [4170]. Claude for Teachers launched [4167]. Gladstone Institutes prize awarded to Trialign [4172]. Google Gemini CLI received multiple fixes [4166, 4165, 4164, 4163, 4162, 4161, 4160, 4159, 4158]. Claude Code desktop app received numerous fixes and features [4157-4129].[46]

No upcoming catalysts are mentioned in the provided claims.