Skip to content
Learn Compare Recipes Ecosystem Source Code Docs

Changelog

OpenCode ships updates frequently. This page is a curated summary; for the authoritative list of every tag, binary, and signed asset, see the OpenCode GitHub releases.

Current release

  • Version: v1.18.32
  • Released: 2026-09-21
  • Highlights: Bedrock image attachments are now only hoisted for Claude, Nova, and Llama 4 models, and Together AI streaming usage reporting is fixed; see the release notes for the full set of changes.

Recent themes

The following themes summarise changes visible to users of the CLI, TUI, Web, Desktop, and SDK. Refer to the linked release notes for the per-tag detail.

  • Performance and stability — compaction now keeps complete recent turns with clearer summaries for smaller models (v1.18.17); automatic session retries are capped with jitter to prevent retry storms (v1.18.17); transient provider and network errors are retried instead of failing (v1.18.14); responses continue when a model reports an unknown finish reason (v1.18.21); message chronology stays correct on import or legacy IDs (v1.18.15); Claude 5 sessions tolerate stale thinking blocks instead of failing after prompt or tool changes (v1.18.26); provider header and streamed chunk timeouts default to five minutes so slow model startups fail less often (v1.18.27); Anthropic thinking block binding is limited to Claude 5.1+ models so older deployments do not reject requests, with a config opt-out to keep the provider default (v1.18.27); Bedrock DeepSeek model IDs — including ARN-based IDs — are preserved so they resolve correctly (v1.18.30); the Azure and OpenAI provider SDKs were updated to pick up compatibility fixes (v1.18.30); ACP session model, effort, mode, and reasoning chunk boundaries are restored when loading, resuming, or forking sessions (v1.18.31); Bedrock image attachments are only hoisted for Claude, Nova, and Llama 4 models (v1.18.32); Together AI streaming usage reporting is fixed (v1.18.32).
  • Provider authentication — Azure providers sign in with Microsoft Entra ID through the Azure CLI instead of requiring an API key (v1.18.24, fixed in v1.18.25), and Azure CLI sign-in now asks for the resource name directly instead of querying Azure management APIs (v1.18.26); xAI login simplified to a single device-code flow that works headless (v1.18.14); GitHub Copilot requests now send the session ID as the interaction header to improve request tracking across a session (v1.18.28); Codex OAuth model filtering recognizes integer GPT versions like gpt-6 (v1.18.29); the TUI shows remote config authentication errors during startup and exits with a failure status (v1.18.31).
  • Cloudflare AI Gateway — native OpenAI and Anthropic passthroughs for gateway models (v1.18.19); routing fixes for third-party providers and Anthropic dashed model IDs (v1.18.23).
  • Config compatibility — V1 reads supported V2 config fields so mixed setups keep working (v1.18.24); unknown top-level config fields are ignored instead of failing parsing (v1.18.16).
  • Desktop — right-to-left layout fixes across tabs, drawers, and titlebar (v1.18.13); archived sessions drop from Home immediately (v1.18.24); project picker and file-search UX fixes (v1.18.16, v1.18.21); OpenCode account device authentication uses the desktop client ID (v1.18.28).
  • Removed — outdated OpenCode Go first-month discount messaging and pricing (v1.18.22); see the Go page for current pricing.
  • Provider coverage — ongoing additions through Models.dev, including refreshed model aliases; the Astra system prompt was added for GPT-6 models (v1.18.30); reasoning effort variants were added for supported GitLab GPT and Claude models (v1.18.30); summarized adaptive thinking is requested for GitHub Copilot models (v1.18.31); Grok 4.7 was added to Zen and Go (v1.18.32); DeepSeek V4.1 Flash documentation was added to Zen (v1.18.32); consult the Models overview for the supported list.
  • Desktop app — the desktop client remains in beta; check the Desktop app page and the official download page for current platform support.
  • Documentation — clearer upgrade notes, refreshed model and Zen documentation, and tighter multi-language coverage.

Breaking changes

Review the upgrade guide before bumping major or minor versions.

  • Configuration key renames when behaviour is replaced rather than extended. Migration hints appear in the affected page, for example Permissions.
  • Deprecated provider models are removed after their announced deprecation date. See the Models overview for the current deprecation list.

Deprecations

  • Long-standing deprecation tables in the documentation (e.g. Models overview, Zen models) list each model, its status, and the announced date. Replace any usage with the provider’s current model alias or a recommended alternative.
  • The desktop app is still in beta; production deployments should track the official download page and the GitHub release feed.

How to upgrade

  1. Read the upgrade guide for the version you are coming from.
  2. Skim the relevant release notes between your current version and the target.
  3. Run opencode upgrade (when available) or follow the installation instructions for your platform.
  4. Verify your configuration by running opencode doctor and your usual smoke-test commands.

See also