Changelog
New updates and improvements to OpenCode
v1.18.25 +1 V1 now reads supported V2 config fields so newer config files keep working in more mixed setups
Improvements
- CoreAzure providers can now sign in with Microsoft Entra ID through the Azure CLI instead of requiring an API key.
Fixes
- CoreBedrock reasoning responses no longer get cached into unreplayable empty messages.
- DesktopArchived sessions disappear from the Home list immediately.@NathanTCode
- CoreFixed Azure authentication so Azure CLI sign-in works without requiring Bun.
v1.18.23 Fixed parent session IDs being sent in request headers for session-aware providers
Fixes
- CoreFixed Cloudflare AI Gateway routing for third-party providers so non-Workers models work through the gateway's REST API.@superhighfives
- CoreFixed Anthropic models through Cloudflare AI Gateway by converting dotted model IDs like
claude-haiku-4.5to the dashed slug Anthropic expects.@superhighfives - TerminalFixed GitHub auth for immutable OIDC subject tokens.
v1.18.22 Removed outdated OpenCode Go first-month discount messaging and pricing
Fixes
- CoreFixed OpenCode device login links when servers return relative verification URLs or use a base path.
- CoreFixed
textVerbositybeing sent to OpenAI-compatible providers that do not support it.@joelstucki-taulia - CoreUpdated the Amazon Bedrock provider for compatibility fixes.
- DesktopKeep model provider headers visible while scrolling the model picker.
v1.18.21 Continue responses when a model reports an unknown finish reason instead of stopping early
Fixes
- CoreRoute Vertex AI
euandusmulti-region Gemini requests through REP endpoints - DesktopKeep file search results visible while the next search is loading
- DesktopRegister the archive session command in both desktop layouts@NathanTCode
- CoreRoute Vertex AI
v1.18.20 Surface failed subagent tool calls with a resumable
task_idFixes
- CoreRetry provider responses that end with
finish_reason: network_error. - CoreRetry more network error variants, including
network-errorandnetwork_error. - CoreSurface resumable subagent failures instead of returning an empty result.
- CorePreserve Cerebras
max_completion_tokenswithout applying an extra output cap. - CoreAnswer permission requests triggered by subagents during
opencode run. - CoreRetry xAI capacity and temporary unavailability stream errors.
- CoreRetry provider responses that end with
v1.18.19 Added native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway models
Improvements
- CoreMatched Codex rate limits more closely to ChatGPT subscription limits.@GameOn223
Fixes 9
- CoreRemoved built-in Qwen sampling defaults that could send unsupported settings.
- CoreProperly show authenticated providers in
/connect.@OpeOginni - CoreIgnore malformed model pricing instead of breaking usage cost calculation.
- CoreFall back when OpenAI websocket requests exceed message size limits.
- CoreForward ChatGPT workspace compute residency to Codex requests.
- CoreUpdated the default Console URL used by account connection flows.
- CoreEnabled web search when using the OpenCode Go provider.
- CorePreserved compatibility with existing v1 databases.
- DesktopKept saved server details editable in the server dialog.
v1.18.18 Select the Kimi system prompt correctly for official Moonshot and Kimi providers
Fixes
- CoreFix xhigh reasoning effort for xai models
v1.18.17 Made session compaction keep complete recent turns and produce clearer summaries for smaller models
Fixes 7
- CoreAdded MERGE Gateway reasoning variants so those model options work correctly.@MatthewFeroz
- CoreCapped automatic session retries and added jitter to reduce repeated retry storms.
- CoreEnabled PDF attachments for GitHub Copilot models that advertise PDF vision support.@stevenao
- CoreApplied the correct sampling defaults to DeepSeek V4 Flash on supported providers.
- CoreRouted all Muse family models to the correct Meta system prompt.@mreso
- DesktopUpdated Chinese developer terminology to use more widely recognized translations.
- DesktopUsed the server's current default model instead of stale local config when choosing the default model.
v1.18.16 Open the project menu with a right-click in Home
Fixes
- CoreIgnore unknown top-level config fields instead of failing config parsing
- CoreRegister projects opened from Home so they are available to the rest of the app
- DesktopFall back to listing and matching local directories when the project picker server does not support search
- DesktopUse 词元 instead of 令牌 for token-related labels in Simplified Chinese@Speechlessmanbilibili
- DesktopKeep the macOS app running after the last window closes and reopen a window when the app is activated
v1.18.15 Added much broader locale coverage across the desktop app
Improvements
- DesktopExport full session transcripts as JSON from the UI.
Fixes 10
- CoreChronological message ordering now stays correct even when imported or legacy message IDs are out of order.
- CoreRevert and fork actions now use real message chronology instead of message ID ordering.
- CoreTruncation cleanup now removes stale files by file timestamp more reliably.
- CoreRepeated compaction now keeps earlier tool-call history in summaries instead of dropping orphaned results.
- CoreBlob-based attachments now load correctly in the web UI.@rexdotsh
- DesktopSession timelines now respect chronological message boundaries more reliably.
- DesktopStored messages now stay ordered by creation time.
- DesktopSession lists now sort by persisted activity time more reliably.
- DesktopLanguage names now stay native in the locale picker.
- DesktopFilled in missing translations across existing locales.