What changed in OpenCode 2, and does it cost more?
OpenCode 2 is free like OpenCode 1: you pay only for the models it calls. Since OpenCode 1.19 (October 11, 2026), V1 no longer updates itself and offers the upgrade with opencode upgrade --major. Your existing opencode.json, custom providers included, keeps working; V1 plugins, server API clients and language-server features do not.
What breaks when you upgrade
| Change | What to do |
|---|---|
| Plugins use a new API | V1 plugins don't run in V2. Port them, or wait for their authors, before you switch. |
| New server API | Scripts and integrations that call OpenCode's server need the new contracts. |
One global cli.json | Terminal settings move from layered tui.json files to one file. OpenCode migrates it for you. |
| No language servers | V2 keeps your lsp settings but runs no language servers and gives the agent no LSP diagnostics. Point it at your lint, typecheck or compiler commands instead. |
| One binary | V1 and V2 both use the opencode command and aren't installed side by side. Remove a package-managed V1 install first; the curl installer replaces it. |
Source: OpenCode's V1 migration guide and OpenCode 1.19.0, checked October 14, 2026.
The changes that affect your bill
| Change | Cost effect |
|---|---|
| Cost in the session view (2.0.29) | Each session shows its context tokens and accumulated cost, so you see a long session growing before the invoice does. |
| Prompt-cache markers for custom providers (2.0.28) | compatibility.cacheFormat sends Anthropic- or Alibaba-style cache markers. On Claude Sonnet 5.5 a cached read costs $0.10 per million tokens instead of $2. |
| Longer stream timeout (2.0.28) | Raised from 5 to 15 minutes, so fewer slow reasoning turns are cut off and rerun. |
| Compaction by token budget | V2 keeps recent context by compaction.keep.tokens. V1's tail_turns and prune are ignored with a warning, so check how much context each turn now resends. |
| ChatGPT sign-in (2.0.27) | GPT-6.1 Sol accepts up to 500,000 input tokens per request on a ChatGPT plan, and bigger requests use your plan's allowance faster. |
| Free models (2.0.26) | Step 5 Preview Free joined OpenCode's free model lineup for a limited time. |
Source: OpenCode changelog, October 8 to 14, 2026. Cache price from Anthropic's pricing page.
Custom providers in OpenCode 2
V2 reads the same config files as V1, and the V1 shape still works, so a provider you added with provider, npm and a base URL keeps running. The native V2 shape is optional: provider becomes providers, npm becomes package, and the endpoint moves to settings.baseURL. Keep each provider entry entirely in one shape, since V2 doesn't mix them inside a single provider.
Two built-in provider IDs merged: azure-cognitive-services is now azure and google-vertex-anthropic is now google-vertex. Update a top-level model that names either one yourself.
Before you run opencode upgrade --major
- Keep a copy of your V1 config. Both versions read the same files, so don't point V1 at a file converted to V2-only fields.
- List your plugins and check each one has a V2 version.
- Start V2 and check models, credentials, agents, permissions and MCP servers before converting anything.
- Open a long session and watch the new cost display for a day, so a change in compaction shows up before the bill does.
The catch
Upgrading costs nothing, but V2 drops V1 plugins and language-server features, and nothing in it caps your bill. A long session on a frontier model still costs whatever the provider charges per token.
Where Standard Compute fits
OpenCode connects to any OpenAI-compatible provider in V1 and V2, so it runs on Standard Compute with base URL https://api.stdcmpt.com/v1 and model standardcompute. Plans start at $19 a month with a $20 compute budget, and requests stop at the budget instead of billing overage. Smart routing sends routine agent steps to efficient models.
FAQ
How do I upgrade OpenCode 1 to OpenCode 2?
Update to OpenCode 1.19 or later and run opencode upgrade --major, or use /update in the terminal client. V1 no longer installs updates automatically. If a package manager installed V1, remove that install first.
Does my opencode.json still work in OpenCode 2?
Yes. V2 reads the same files and normalizes supported V1 fields without rewriting them. Plugins, the server API and terminal settings are the three intentional breaking changes, and some fields such as compaction.tail_turns are ignored with a warning.
Does OpenCode 2 cost money?
No. OpenCode is still free; you pay for the models it calls, through your own API key, a ChatGPT sign-in, OpenCode's Go plans or a flat-rate API.
Can OpenCode 2 use a custom OpenAI-compatible endpoint?
Yes. Keep your V1 provider entry, or use the native shape with providers, package @opencode/ai/providers/openai-compatible and settings.baseURL. Standard Compute works at https://api.stdcmpt.com/v1 with model standardcompute.