Repository navigation
Go gateway: SSE streaming requests over HTTP/1.1 return HTTP 500 (Endpoint is unavailable) #42843
Description
Description
Summary
Streaming (SSE) chat completion requests to the opencode-go gateway endpoint https://opencode.ai/zen/go/v1/chat/completions return HTTP 500 with body {"type":"error","error":{"type":"error","message":"Internal server error"}} when sent over HTTP/1.1. The same request over HTTP/2 succeeds (HTTP 200 with normal SSE chunks). Non-streaming requests over HTTP/1.1 also succeed.
This surfaces in OpenChamber/openCode sessions as AI_APICallError: Upstream request failed: Endpoint is unavailable. for provider opencode-go (e.g. deepseek-v4-flash), in recurring bursts (retries all fail for several minutes, then recover).
Evidence
- HTTP/1.1 + stream:true -> always HTTP 500 (reproduced 10+ times)
- HTTP/1.1 non-streaming: HTTP 200 (5/5)
- HTTP/2 (node http2) + stream:true -> HTTP 200, normal SSE stream
- opencode CLI 1.18.18 (bun, HTTP/2) with
opencode-go/deepseek-v4-flashworks fine - Request body/headers captured from the working client and replayed via curl over HTTP/1.1 still 500s:
{"model":"deepseek-v4-flash","max_tokens":32000,"top_p":0.95,"messages":[...],"tool_choice":"auto","stream":true,"stream_options":{"include_usage":true}}
Expected
Streaming over HTTP/1.1 should work like any OpenAI-compatible endpoint (other providers do). Or at minimum return a meaningful status/error instead of a generic 500.
Plugins
None
OpenCode version
1.18.18 (bundled with OpenChamber desktop)
Steps to reproduce
- curl.exe -X POST "https://opencode.ai/zen/go/v1/chat/completions" -H "Authorization: Bearer " -H "Content-Type: application/json" -d '{"model":"deepseek-v4-flash","messages":[{"role":"user","content":"hi"}],"stream":true}' -> HTTP 500 {"type":"error","error":{"type":"error","message":"Internal server error"}}
- Same request without "stream":true -> HTTP 200, normal completion
- Same request over HTTP/2 (e.g. node http2 module, or bun fetch used by the opencode CLI) -> HTTP 200, normal SSE stream
Screenshot and/or share link
No response
Operating System
Windows 11
Terminal
OpenChamber desktop (bundled opencode-cli)
Activity
github-actions commentedon Aug 16, 2026
This issue might be a duplicate of existing issues. Please check:
Endpoint is unavailableerrors on the same endpoint with analysis of the Cloudflare/upstream relationshipUpstream request failed: Endpoint is unavailableon v1.18.18If your issue is specifically about HTTP/1.1 streaming being broken while HTTP/2 streaming works (i.e. a protocol-level difference rather than intermittent unavailability), it may be distinct from the above — the HTTP/1.1 vs HTTP/2 reproduction steps make this a useful data point worth keeping separate if that distinction holds up.
118Gavin commentedon Aug 16, 2026
Thanks for the duplicate check. This one is a distinct data point worth keeping: it isolates a deterministic protocol-level difference (HTTP/1.1 + SSE -> guaranteed 500; HTTP/2 + SSE -> 200, 100% reproducible in the same minute from the same machine/key/network), not intermittent unavailability.
Comparison with the referenced issues:
Endpoint is unavailablein ~5-10 min bursts, e.g. 17:32-18:10 and 01:27-01:34 UTC), so the outage cluster likely explains the bursts users see in OpenChamber sessions. The h1.1-streaming 500 is an additional deterministic bug on top of that.Reproduction is one curl call, so it's cheap for the assignee to verify:
curl -X POST https://opencode.ai/zen/go/v1/chat/completions -H "Authorization: Bearer <key>" -d '{"model":"deepseek-v4-flash","messages":[{"role":"user","content":"hi"}],"stream":true}'-> HTTP 500; same request via HTTP/2 (node http2) -> 200.Also worth noting for the outage angle: #40480 mentions an "OpenCode Go China-hosted model opt-in" console setting — my traffic is from a China IP; if that opt-in routes to different upstreams, it may be relevant to the burst pattern. Happy to close as duplicate of #40480 if maintainers prefer, but the h1.1/h2 repro is unique to this report.
118Gavin commentedon Aug 16, 2026
Additional data point for the assignee: the account in question has the "Enable models hosted in China" opt-in enabled the whole time (and the failure is
Endpoint is unavailable/500, never the 403 RegionError from the opt-in gate). So this is not the opt-in/RegionError class of issue (#39845) — it's the gateway's streaming path failing even for China-hosted routing. The h1.1-vs-h2 repro in the body stands as-is.thdxr commentedon Oct 11, 2026
Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (
opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊