OpenCode v1.4.7: Reasoning Models & Auth Context

TL;DR

• GitHub Copilot gpt-5-mini now uses low reasoning effort for better compatibility
• Auth context persists across workspace sessions
• Cloudflare AI Gateway fix stops GPT-5 and o-series requests from failing
• Claude Opus 4.7 gains xhigh adaptive reasoning support

New

• Claude Opus 4.7 xhigh adaptive reasoning — Adds support for the highest reasoning effort level on Opus 4.7 models.
• Claude Opus 4.7 summarized thinking — Thinking output now displays in summarized form by default.
• Session replay to workspaces — Sessions can now replay into another workspace in batches.
• Auth context in workspaces — Provider sign-in now carries across workspace sessions.

Fixed

• GitHub Copilot gpt-5-mini reasoning — Now uses low reasoning effort to improve request compatibility.
• Cloudflare AI Gateway OpenAI models — Drops max_tokens for reasoning models so GPT-5 and o-series requests stop failing.
• Azure model defaults — Models now default store=true, fixing requests that require stored responses.
• TUI file paste — Pasting files or large text no longer inserts content twice.
• TUI agent override — --agent flag on command line is no longer overwritten by saved session agent.
• TUI plugin loading — Plugins now load against the correct project when multiple directories are open.
• Bash tool memory — Uses less memory when handling large command output.
• Workspace sync writes — Experimental workspaces now wait for sync to finish before returning writes, reducing stale reads.
• Provider 5xx retries — Sessions now retry provider errors even when the provider SDK doesn't mark them retryable.
• LSP/MCP status responses — Empty responses no longer break TUI sync state.

Other

• Desktop build badges — Beta or Dev badges now display in title bar when applicable.

Update with: npm install opencode@latest or check the release page for platform-specific instructions.

Source: OpenCode