OpenCode 2 v2.0.24: native Vercel AI Gateway provider, AI SDK v6 by default#

Published October 6, 2026 by Alex Sorokoletov

Part of the OpenCode 2 Version Tracker series. | OpenCode on GitHub | OpenCode 2 docs

OpenCode 2 v2.0.24 follows v2.0.23 with a provider-weighted set of changes: Vercel AI Gateway becomes a native provider with its language models in the catalog, provider integrations install the AI SDK v6 packages by default, and the catalog drops non-text models while routing completions-shaped Azure entries through chat.

Providers#

Vercel AI Gateway is now a native provider. The gateway fronts many model vendors behind one endpoint and credential, and OpenCode registers its language models in the catalog (#52643) and wires the provider directly (#53432).^[1] Alongside it, OpenCode installs the AI SDK v6 provider packages by default (#53287), the version of the SDK its provider plugins build on.

The rest of the provider work is catalog correctness.

Sign-in labels for OpenAI accounts moved around. Legacy OpenAI OAuth methods are relabeled Codex (#53467), the ChatGPT OAuth method labels are clarified (#53451), and model-list syncing over the /models API is paused for ChatGPT sign-in (#53466).^[2] On the MCP side, a failed OAuth token exchange now reports on the callback page instead of failing without a message (#53378).

Interface#

The terminal interface lists sessions awaiting input at the top of the open menu, so a run blocked on a question is visible without scrolling (#53435). The subagent activity filter persists across navigation (#53318), and mermaid flowcharts render without crossing branches (#53325).

The desktop app clarifies which work is running in session headers (#53332) and distinguishes queued from pending steer messages (#53338). In the timeline, errors and grouped updates get their own spacing (#52876). A core fix stops the instruction-file walk at case-variant Windows project roots, so instruction collection no longer climbs past the repository on Windows (#53349).

What this tells us#

Vercel AI Gateway as a native provider widens the ways you bring a model to OpenCode. The gateway routes through a single endpoint that fronts many vendors, and it joins direct providers already in the catalog: Anthropic, OpenAI, Google Vertex, Azure, GitLab Duo, and local runtimes like Ollama, LM Studio, and vLLM. Claude Code, Codex CLI, and Gemini CLI each anchor to one vendor's models; OpenCode keeps expanding the set of entry points, whether a direct key, an OAuth plan, or a gateway.

The release also ships AI SDK v6 providers by default (#53287) and four provider-specific catalog fixes: Azure completions routing (#53395), non-text model exclusion (#53295), GitLab Duo reasoning variants (#51082), and Vertex effort updates (#53289).

The split between terminal and desktop work tracks OpenCode 2's multi-surface shape. Open-menu surfacing and subagent filtering land in the terminal interface; session headers and steer-message states land in the desktop app. The docs describe a terminal interface, a desktop application, and a web surface over one background server, a heavier footprint than the single-process runs of Claude Code, Codex CLI, and Gemini CLI, and this release spends on both front ends in the same range.

All trademarks and software referenced belong to their respective owners.

Sources#

1. Vercel AI Gateway and AI SDK v6: #52643, #53432, #53287
2. OpenAI sign-in labels: #53467, #53451, #53466