OpenCode v1.1.48: Provider Fixes & Model Config

TL;DR

• Custom models file path now configurable via OPENCODE_MODELS_PATH env var
• Copilot provider now properly handles reasoning tokens
• Skills now invokable as slash commands in TUI
• Various provider fixes for OpenAI-compatible APIs and Mistral/Devstral ordering

New

• Custom models path configuration — specify your models file location via OPENCODE_MODELS_PATH environment variable
• Skills as slash commands — invoke skills directly as slash commands in the TUI for faster workflows
• Copilot reasoning token handling — dedicated Copilot provider now properly processes reasoning tokens
• OPENCODE_MODELS_URL in build — build process now respects OPENCODE_MODELS_URL environment variable

Fixed

• OpenAI-compatible API parameter naming — thinking parameter now uses snake_case for proper compatibility
• Mistral/Devstral ordering — ordering fixes now consistently applied to both providers
• Bash permissions handling — fixed treesitter node handling for redirected statements in bash
• Symbolic link following — grep and ripgrep operations no longer follow symbolic links by default
• Model configuration validation — ensures models configuration is not empty before loading
• Google Vertex Anthropic provider — removed special case handling in response generation
• Test environment isolation — parallel test runs no longer contaminate environment variables
• Chat model verbosity — chat models now excluded from textVerbosity setting

Desktop

• UI reverts — transitions, spacing, scroll fade, and prompt area updates reverted
• Test coverage — added session actions and project tests

Update via: npm install opencode@1.1.48 or check the release page for your platform.

Source: OpenCode