Repository navigation
DeepSeek V4 Flash (New) route hangs - no response from zen gateway #40479
Description
Description
The deepseek-v4-flash model (marketed as "DeepSeek V4 Flash (New)") on the opencode zen gateway fails to respond.
Endpoint
POST https://opencode.ai/zen/go/v1/chat/completions
Authorization: Bearer $OPENCODE_API_KEY
Symptoms
- Without streaming: 0 bytes received, times out after 90s
- With streaming: Only SSE
: keep-alivepings, no actual content arrives - With/without max_tokens: same result
- Billing: user is paying and has balance available
Reproduction
curl -sS -m 60 "https://opencode.ai/zen/go/v1/chat/completions" \
-H "Authorization: Bearer $OPENCODE_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"model":"deepseek-v4-flash","messages":[{"role":"user","content":"hello"}]}'
# Result: timeout, 0 bytesComparison
Other models via the same endpoint/key work fine:
qwen3.8-max→ 2sglm-5.2→ 1.6sdeepseek-v4-pro→ 13s
Environment
- opencode v1.18.13
- Linux (Ubuntu 20.04)
- git v2.50.1
Expected
deepseek-v4-flash should respond like other models.
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Aug 4, 2026
This issue might be a duplicate of existing issues. Please check:
deepseek-v4-flashonopencode.ai/zen/go/v1drops connection before any HTTP response whiledeepseek-v4-proon the same endpoint/key works fine; includes curl reproduction.DeepSeek V4 Flashstopped working (both free and paid tiers),deepseek-v4-prostill works — same provider, same version (v1.18.13).Deepseek V4 Flash Freenot working — another report of the same hang/no-response pattern.If your issue is the same underlying server-side routing problem with the flash lane on the zen/go backend, please add a reaction or comment to one of those existing issues rather than opening a new one.
ci: switch review bot to OpenCode Zen free tier (keyless) (#219)
KarpovVolodymyr commentedon Aug 31, 2026
Any plans to fix it soon?
Having the same issue
Vlyn commentedon Oct 7, 2026
I'm using
/zen/go/v1/chat/completionswithopencode-go/deepseek-v4.1-flashas theclassifier model for a coding agent (pi, Windows 11). Requests are tiny, one-token answers.
Symptom: the request never returns a single byte. No HTTP status, no error, no stream
data, my client's own 20s deadline aborts it. It's seemingly random, working fine one second,
then no response the next.
What makes me think it's routing/edge rather than request shape: byte-identical bodies
succeeded in 1.0–2.5s on both sides of the failure (1.6s just before, 2.5s just after),
and one other call in the same run took 18.7s, just under my deadline. So it's a latency
tail with a cutoff, not a dead lane, unlike #40465, where flash fails every time.
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.