What changed in the major AI coding CLIs?

OpenAI released Codex v0.144.3 with remote control via ChatGPT app, multi-agent handoffs, DigitalOcean integration, and indexed web search [5497, 5496]. Codex Mac app is overloading SSDs, causing permanent damage, with multiple user reports of fried drives [5499]. Google released Gemma 4, a local AI model that runs up to 3x faster with MTP drafters [5494]. Kimi K3, a 2.8 trillion parameter open-weight model, benchmarks near Fable 5 and GPT 5.6, completing 122 coding tasks autonomously [5491]. Claude Final Boss, an open-source agentic OS for Claude Code, has 200K+ GitHub stars, featuring 63 agents, 250 skills, and 80 commands [5490]. A GitHub repo with 26k stars enables Claude Code to use free models like DeepSeek and Kimi, causing 20,000 users to switch [5488]. Anthropic released an official plugin claude-code-setup that analyzes projects and recommends automations [5487]. Anthropic released Frontend Design, an official skill for Claude Code with 161,000 GitHub stars [5486]. Moonshot AI released Kimi Code CLI, an open-source MIT-licensed CLI that clones Claude Code with additional features [5485]. Anthropic standardizes Claude Fable 5 access at 50% usage for plans that use it most intensively [5482]. Anthropic launches Claude for Teachers, offering free premium Claude capabilities to verified K-12 educators in the US [5481]. Google Gemini CLI received multiple fixes: strip thoughts from scrubbed history turns [5480], preserve escape sequences in string literals [5479], make ~/.gitconfig read-only in macOS sandbox [5478], egress Cloud Run service skeleton [5477], resolve symbolic link directory escape in memory import processor [5476], implement Cloud Run webhook ingestion service [5475], resolve defensive path resolution for at-reference files [5474], enforce case-insensitive sensitive path blocklist [5473], fix no_proxy test [5472]. Codex desktop app received fixes: removed obsolete Codex workaround [5471], preserved composer caret after request updates [5470], resynced timeline correctly after reconnecting [5469], loaded more timeline history at once [5468], fixed project picker positioning crash [5467], kept permission auto-accept state separate per server [5466], fixed connected project picker positioning [5465], fixed spacing between model provider sections in Settings [5464], fix(core): tolerate AlreadyExists in FSUtil.ensureDir [5463], hid drawer close button on Windows [5462], stopped subagents from launching nested subagents by default with configurable subagent_depth limit [5461], home page command palette search can find and open sessions [5460], fixed custom agent selector visibility [5459], fixed startup readiness including WSL server loading [5458], improved dangerous-command detection [5457], fixed plan approvals without edits being labeled "(edited by user)" [5456], fixed hook callback timeout misreported as user rejection [5455], fixed session crash when tool result renderer returned bigint or plain text [5454], fixed paste markers leaking into external editors [5453], fixed rendered text fragment leaking into crash telemetry [5452], fixed ultracode keyword opt-in firing on non-human-originated input [5451], added startup warning for Write/NotebookEdit/Glob permission rules [5450], added live elapsed-time counter to collapsed tool summary line [5449], fixed edits that leave input as "?" being silently swallowed [5448], fixed file upload validation for DOS device suffixes and multiple hard links [5447], fixed Claude Code on Vertex and Bedrock attempting default Opus model at startup [5446], fixed auto mode overriding PreToolUse hook's ask decision for unsandboxed Bash [5445], fixed permission previews not neutralizing bidirectional-override characters [5444], fixed shell mode not executing commands containing file paths while autocomplete popup was open [5443], fixed SIGTERM during Bash tool orphaning process tree [5442], fixed continue:false hook's halt being dropped [5441], fixed plan mode auto-running file-modifying Bash commands without permission prompt [5440], added session-wide limit on WebSearch tool calls (default 200) [5439], added message.uuid, client_request_id, and tool_source to OpenTelemetry log events [5438], added ISO modified timestamp to memory file frontmatter [5437], fixed Bash permission checks for help/man commands [5436], fixed Bash permission checks for zsh variable subscripts [5435], fixed Bash permission checks for very long commands [5434], fixed Bash permission checks for file-descriptor redirect forms [5433], fixed permission-check bypass in Windows PowerShell 5.1 [5432], fixed single-segment dir/** allow rules [5431].[62]

OpenAI released Codex v0.144.3 with remote control via ChatGPT app, multi-agent handoffs, DigitalOcean integration, and indexed web search [5497, 5496]. Codex Mac app is overloading SSDs, causing permanent damage [5499]. Google released Gemma 4 [5494]. Kimi K3, a 2.8 trillion parameter open-weight model, was released [5491]. Claude Final Boss, an open-source agentic OS for Claude Code, reached 200K+ GitHub stars [5490]. A GitHub repo with 26k stars enables Claude Code to use free models, causing 20,000 users to switch [5488]. Anthropic released claude-code-setup plugin [5487]. Anthropic released Frontend Design skill [5486]. Moonshot AI released Kimi Code CLI [5485]. Anthropic standardizes Fable 5 access at 50% usage [5482]. Anthropic launches Claude for Teachers [5481]. Google Gemini CLI received multiple fixes [5480-5472]. Codex desktop app received multiple fixes [5471-5431].[62]

No upcoming catalysts are mentioned in the provided claims.