πŸ”₯ Claude Code β€” v2.1.154 (Today)

Anthropic released Claude Code 2.1.154 today, a major update centered around the new Claude Opus 4.8 model.

New Features

  • Opus 4.8 is here β€” now defaults to high effort for deeper analysis out of the box
  • /effort xhigh β€” a new effort level for your toughest coding tasks
  • Dynamic workflows β€” ask Claude to create a workflow and it orchestrates work across tens to hundreds of agents in the background for larger, more complex tasks. Run /agents to view your runs
  • Fast mode on Opus 4.8 β€” now available at a fraction of the previous cost: 2x the standard rate for 2.5x the speed
  • The lean system prompt is now the default for all models except Haiku, Sonnet, and Opus 4.7 and earlier
  • /approve now runs a cleanup-only review (reuse, simplification, efficiency, altitude) instead of the full bug-hunting review
  • /bg β€” run a shell command as a background session you can attach to and detach from
  • Chrome browser selection β€” pick which connected browser to use via settings β†’ "Select browser…", or in-chat when a browser action runs with multiple connected
  • Plugins can now declare defaultEnabled: false in their manifest or a marketplace entry; enable them with settings or the marketplace UI
  • Plugin Discover tab now pins plugins whose relevance signals match the current directory with a "suggested for this directory" annotation
  • Streaming tool execution is now always enabled, including when telemetry is disabled or on Bedrock/Vertex/Foundry
  • Stdio MCP server subprocesses now receive CLAUDE_CODE_VERSION and CLAUDE_CODE_ENVIRONMENT in their environment
  • /agents to open the agents view now works on Bedrock, Vertex, Foundry, and with telemetry disabled

Safety & Improvements

  • Slider labels renamed from "Speed"/"Intelligence" to "Faster"/"Smarter" for clarity
  • Improved auto-mode classifier's detection of data exfiltration, particularly bulk transfers of repository contents
  • Claude now reserves multiple-choice questions for decisions it genuinely cannot make itself
  • Deprecated /turbo (will be removed 06/01). Switch to fast mode via model settings instead

Bug Fixes

  • Fixed dangerous path blocking when directory has a trailing slash (safety fix)
  • Numerous fixes across background sessions, worktrees, VS Code, and Windows

Claude API Platform

  • Claude Opus 4.8 is now generally available on the API
  • Compliance API integrations announced β€” stronger governance with security and compliance tools
  • Increased throughput for Claude Opus 4.8 API endpoints, resolving minute-level bottlenecks for agents and batch jobs

Links: Changelog | GitHub | Platform Release Notes

---

OpenAI Codex CLI β€” v0.135.0 (Today)

OpenAI released Codex CLI 0.135.0 just hours ago.

New Features

  • Richer diagnostics β€” reports environment, Git, terminal, app-server, and thread inventory diagnostics for support cases
  • Vim mode gained text-object editing, improved word/line-end behavior, and a configurable interrupt-turn binding
  • /permissions now understands named permission profiles and displays configured custom profiles
  • Bundled patched zsh helper now discovered across supported macOS and Linux targets
  • Python SDK exposes friendly presets for thread and turn APIs

Bug Fixes

  • Markdown tables and multiline lists render more readably in TUI with better column sizing
  • TUI output stability on macOS and Zellij β€” fixed stderr/composer corruption and raw-output overlap
  • Slash-command completion preserves existing draft text for inline arguments
  • Older tmux/iTerm control-mode sessions no longer lose key handling
  • App mentions now exclude inaccessible or disabled apps
  • Resume flows include non-interactive exec sessions and honor cwd overrides for idle cached threads

Documentation

  • Updated Python SDK docs and notebook content to use the new sandbox preset API
  • Clarified image-viewing tool detail behavior

Links: Changelog | GitHub Releases | Install: npm install -g @openai/[email protected]

---

OpenCode β€” v1.15.12 (Today)

OpenCode shipped v1.15.12 hours ago β€” the latest release of the community-driven open-source coding agent.

Core

  • ACP integrations can now send prompts, slash commands, and usage updates through acp-next
  • Added WebSocket transport for OpenAI responses on supported channels (set OPENCODE_EXPERIMENTAL_WEBSOCKETS=true)
  • Workspace management dialog added to TUI

Bug Fixes

  • Enabled adaptive reasoning controls for Anthropic Opus 4.7+ models
  • Fixed colons in passwords (#29180)
  • Improved acp-next model and config switch startup time and first-session startup
  • OpenAI WebSocket: kept timeouts active, retried failed streams before fallback, handled permission prompts correctly
  • Fixed remote workspace request body forwarding
  • Custom base URLs for OpenAI WebSocket responses
  • Session navigation now works while prompt modes are open
  • Restored the thinking spinner
  • Surfaced subagent retry status
  • Fixed opening editors from non-Git project paths
  • Desktop: improved home empty state, fixed V2 font usage, fixed tab close buttons

Links: GitHub Releases | Website | GitHub

---

Competitive Landscape Notes

  • All three tools shipped major updates today β€” Opus 4.8 is the big driver, with Claude Code getting the full suite first, OpenCode patching support for the model's adaptive reasoning, and Codex CLI leaning into diagnostics and Vim mode polish
  • Claude Code's dynamic workflows (orchestrating tens-to-hundreds of background agents) is a significant differentiator that neither Codex nor OpenCode has fully matched yet
  • OpenCode continues rapid iteration (~167k GitHub stars) with strong WebSocket and ACP integration
  • OpenAI Codex CLI remains popular (86.6k stars) with steady improvement on TUX stability and developer ergonomics

---

AI Coding Intelligence β€’ Leaniverse News