Repository navigation
LongCat 2.5 Preview Free: 500 before compaction, then opaque 429 #52288
Description
Description
OpenCode Desktop intermittently failed on longcat-2.5-preview-free in one session. A request returned HTTP 500 inference_failed twice. After I compacted the session, the next request was much smaller but returned HTTP 429 inference_failed immediately. I cannot tell from the console record whether the 429 was an expected usage limit or an incorrect rejection.
Both requests used the OpenCode provider, product standard, streaming openai-chat, and POST /inference/openai/v1/chat/completions.
- Sep 30, 2026, 14:46:48 (Egypt local time): HTTP 500, request ID
a5c02226-b621-44ad-8a0f-77b7757c92c8, 1,330,241-byte request. Two attempts failed after 1.99 s and 1.84 s; total 4.03 s. - After
/compact, Sep 30, 2026, 15:34:26: HTTP 429, request IDa4ef3157-3a1c-4639-9ef4-f4a07630fede, 208,992-byte request. Rejected in 160 ms. The record provided no more specific limit reason or reset time.
Because the compacted request was rate limited, I have not yet confirmed whether compaction prevents the 500. Please check the two request IDs for the cause of each response and whether the 429 was the intended limit behavior.
Plugins
Not confirmed.
OpenCode version
Desktop 2.0.20
Steps to reproduce
- Use
longcat-2.5-preview-freein a long OpenCode Desktop session. - Continue the conversation; the larger request may return HTTP 500
inference_failed(intermittent). - Run
/compactand retry in the same session. In this instance, the next request returned HTTP 429inference_failed.
Operating System
Windows 11 Pro
Terminal
OpenCode Desktop
Related issues
#41236 reports intermittent free-model failures. #52174 concerns inference_failed on a large request, but on another model with HTTP 400. #51682 concerns Go usage-limit blocking; these requests were product standard.
Activity
github-actions commentedon Sep 30, 2026
This issue might be a duplicate of existing issues. Please check:
inference_failedon oversized payloads with max reasoning effort #52174: reportsinference_failedon a large OpenCode gateway request (different model, product, and HTTP status).IbrahimHafez1 commentedon Sep 30, 2026
Additional evidence from the OpenCode Console log (Sep 30, 2026, Egypt local time): with filters User = my account and Status = Failed, the first results page contains 50 requests, all for
longcat-2.5-preview-free, all showing HTTP 500inference_failed, spanning 14:50:13–15:30:55. The Older button is enabled, so 50 is a lower bound, not a total. These are in addition to the 14:46:48 request already in the issue.Representative request IDs across that period:
d353b267-ac05-4a0f-9cf4-8695d304f4c5(500, 4.19 s)5beab106-7304-4a10-ae48-598a40cdae03(500, 4.63 s)06baf261-7ccb-4938-bcd3-d17439ce8003(500, 4.52 s)The first results page contains no other model or HTTP status. This was a repeated failure over roughly 40 minutes, not a single transient 500. I have not established whether all requests had similar payload sizes.
omani commentedon Sep 30, 2026
ikr, this is a shitshow.
omani commentedon Sep 30, 2026
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.