OpenCode v1.18.32 adds TypeSafe AI Jev structured decision models
OpenCode's changelog shows Zen now supports TypeSafe AI Jev (jev-1.13 and a free version) for parallel yes/no, multiple-choice and rubric questions, and adds models including Grok 4.7.
𝙊𝙥𝙚𝙣𝘾𝙤𝙙𝙚 v1.18.32 released. TL;DR: Zen adds TypeSafe AI Jev and Grok 4.7, Bedrock hoists tool images for non-multimodal models, Core fixes Node package entrypoint resolution, and Data sync gains paginated retries. 𝗭𝗲𝗻 • Added support for TypeSafe AI Jev structured decision models (jev-1.13 and jev-1.13-free) via the SystemOne API format, supporting parallel yes/no, multiple-choice, and rubric-based questions. • Added Grok 4.7 to Zen and Go catalogs with zero data retention compliance. • Added MiMo-V2.6-Pro and MiMo-V2.6-Flash to Zen and Go catalogs. 𝗣𝗿𝗼𝘃𝗶𝗱𝗲𝗿𝘀 • Fixed Amazon Bedrock tool result formatting to hoist image attachments into separate user messages for all models except Claude, Nova, and Llama 4. • Updated Together AI integration so streaming responses report token usage. 𝗖𝗼𝗻𝘀𝗼𝗹𝗲 Fixed • Fixed inference proxy requests to authenticate with Cloudflare Access service credentials and forward client IPs via x-zen-ip headers. • Fixed routing for new Console API keys to forward directly to inference without requiring legacy workspace database lookups. Changed • Changed desktop download routes to redirect directly to v2 release API assets. • Changed login redirection to send users without unmigrated workspaces directly to the v2 Console login. 𝗖𝗼𝗿𝗲 • Fixed npm package entrypoint resolution under Node runtimes to resolve to entrypoint file URLs, preventing directory import errors when loading packages in the desktop sidecar. • Fixed an internal module import cycle between filesystem search and filesystem schema definitions. 𝗗𝗮𝘁𝗮 • Fixed R2 SQL stats synchronization to cursor-paginate aggregate queries, avoiding the 10,000-row ceiling without truncating high-cardinality results. • Fixed transient R2 SQL query timeouts and rate limits by retrying failed pages with exponential backoff. • Fixed model name normalization to clamp identifiers longer than 256 characters to unknown to prevent aggregation failures. No noticeable bundle change Compare: https://github.com/anomalyco/opencode/compare/v1.18.31...v1.18.32