OpenCode Desktop is now built around tabs — start a new session in a tab, or open an existing one
OpenCode shipped a major desktop redesign with tabbed sessions, plus a flurry of v1.18.x updates. Here is what changed and what it means for developers using OpenCode as their AI coding agent.
OpenCode Desktop is now built around tabs — start a new session in a tab, or open an existing one
OpenCode Desktop is now organized around tabs instead of a single-session shell. Developers can start a new agent session in one tab, reopen an existing session from another project, and switch between them without losing state. Palette search also finds sessions that are already open.
Each tab can point to a different project directory, which makes the desktop application more practical for developers who move between repositories during the day. The familiar tab model also reduces the need to open multiple application windows just to preserve separate agent contexts.
Git worktrees are not yet supported in the new tab interface. Developers who need parallel checkouts can temporarily switch to the previous layout in Settings or continue using the terminal TUI until worktree support reaches the desktop tabs.
OpenCode v1.18.3 added an Up Arrow shortcut for closing the subagent picker and fixed several desktop issues, including home-page scrolling, sticky session headers, WSL readiness, and custom-agent selector visibility.
Version 1.18.2 changed subagent behavior by preventing nested subagents by default. Teams that need deeper delegation can configure a subagent_depth limit. The release also improved default reasoning depth for Meta models and added Mod+N as another shortcut for opening a new tab.
The earlier v1.18.0 and v1.18.1 releases focused mainly on bug fixes. Nearby v1.17.x updates added Azure AI support for GPT-5.6, OpenAI pro reasoning mode, OAuth for Luna Responses Lite, and Codex context limits for GPT-5.6 over OAuth.
GPT-5.6 support improved across OpenCode's provider integrations. Azure AI can now use GPT-5.6 models, Codex context limits apply correctly to GPT-5.6 Sol over OAuth, and OpenAI pro reasoning mode is available where the provider supports it.
For xAI, OpenCode now exposes reasoning-effort variants for Grok models, improves prompt-cache routing, and adds PDF support for Responses models. Response storage is disabled by default for xAI requests. Meta model handling also received better reasoning-variant detection, provider request logic, and a model-specific system prompt for Meta Muse Spark.
OpenCode enabled adaptive thinking for Claude Sonnet 5 and improved MCP authentication behavior. MCP servers can reconnect after OAuth even when previously disabled, refresh tokens are supported, and OpenCode prefers MCP content when a response contains both content and structured output.
GitHub Copilot integrations no longer crash or display invalid pricing when a model reports a zero billing batch size, and stale response item IDs are no longer replayed in follow-up requests. OpenAI-compatible reasoning models are also forced into reasoning mode so custom deployments apply their reasoning settings consistently.
Together, these changes make OpenCode more reliable across a mixed provider setup. The desktop redesign is the visible update, but the provider and reasoning fixes are likely to matter more for teams that switch between GPT, Claude, Grok, Meta, and compatible third-party endpoints.
APItokendeal provides OpenAI-compatible API access for coding agents like OpenCode, Claude Code, Cline, and Codex. One API key unlocks GPT, Claude, DeepSeek, Gemini, and more with prepaid token packs.