Repository navigation
All opencode/* free models return Upstream request failed: Model is unavailable (gateway-wide) #50282
Description
Summary
As of 2026-09-21, every opencode/* free model returns an upstream 400 server_error: Upstream request failed: Model is unavailable. This is a gateway-wide outage, not a per-model routing bug — opencode/big-pickle, opencode/mimo-v2.5-free and opencode/deepseek-v4-flash-free all fail identically.
Repro
Direct call to the zen endpoint with a valid API token (token is accepted — no 401):
curl https://opencode.ai/zen/v1/chat/completions \
-H "Authorization: Bearer sk-..." \
-H "Content-Type: application/json" \
-d '{"model":"opencode/big-pickle","messages":[{"role":"user","content":"ping"}],"max_tokens":8}'
=> HTTP 400 {"error":{"type":"server_error","message":"Upstream request failed: Model is unavailable."}}
Tested models (all identical 400):
opencode/big-pickleopencode/mimo-v2.5-freeopencode/deepseek-v4-flash-free
Notes
- Response header shows
Server: cloudflare/cf-placement: remote-ORDand includesx-opencode-log-id: 955ca27a-.... - Context: this follows the long-running
big-picklerouting issue (Big Pickle (big-pickle) model returns AI_APICallError - was working before #28141). Today it escalated from intermittent to a total outage of the free tier. - Restoring the
opencodecredential inauth.jsonfixed the earlierInvalid credential401, confirming the token itself is valid — the failures are purely upstream. - My DeepSeek
auth.jsonkey is also rejected by DeepSeek's own API as invalid, so there is no local fallback either.
Expected
Free models should respond as they did before (they worked fine as of 2026-09-19) or the gateway should surface a clearer maintenance/outage status instead of a per-request server_error.
Log id: 955ca27a-f0b8-4f17-... (request above, 2026-09-21 ~04:49 UTC)
Activity
geraistec commentedon Sep 21, 2026
Same problem: #50084
marcelloinfoweb commentedon Sep 24, 2026
This is abuse!!!
mgajda commentedon Oct 18, 2026
Frequency data, in case it helps size the gateway-side fix.
Over 2026-09-24 → 2026-10-16: 56 of 790 ERROR lines (7.1%) were
Upstream request failed: Model is unavailable.— 48 asmessage="stream error", 8 asmessage="process".This is the third most frequent failure I recorded, and it's notable that the message carries no
model or provider identity at all, so there is no way to tell from the log which models are
affected.
For context on the neighbouring class,
Endpoint is unavailableis larger still — 153 lines (19.4%),though that one appears to have several distinct causes tracked separately (#43477, #44850, #53120).
Context: #35325, #36582.
neriousy commentedon Oct 21, 2026
Thanks for the report, and sorry for the disruption. This was a temporary problem with the hosted service or an upstream model provider, and it has since recovered. You can see live status and past incidents at https://status.opencode.ai. If you still see this now, please open a new issue with the time (with timezone), model, and request ID from the console logs.