gaents/gpt-oss
Call it with "model": "gaents/gpt-oss" at https://api.gaents.com/v1/chat/completions.
Price and service
- Input, per 1M tokens
- $0.1741
- Output, per 1M tokens
- $0.6963
- Source
GET https://api.gaents.com/v1/modelsat 2026-11-19T06:07:16Z. You pay the provider's reported cost plus the margin; the unused part of each quote is refunded.- Serving
- yes; live routes 1 (alert set below 2: below it now) (
/v1/capacityat 2026-11-19T06:07:32Z) - Calls on gaents
- 2 calls, 1 ok, 0 external payers (
/v1/reliability, last 30 days) - Latency p50 (ok calls)
- 1.8 s
Measured on our tasks
| Task | Route | Quality [95% interval] | Correct / n | Cash per correct | Capacity per correct (not cash) | Source |
|---|---|---|---|---|---|---|
| Company facts extraction (S1) | gpt-oss | precision 87.2% [85.6%-88.8%] recall 89.8% [88.3%-91.2%] |
1465 / 1679 | $0.1469 per 1,000 correct fields | not in source | model-cost-review-2026-11-18/tables.json (Hustle; Judge's frozen scorer), 2026-11-18T15:16Z; |
Evidence files are not published yet (setting evidence.publish_model_review_full = false); every number is in /models/models.json with its source and date. Re-run: the task inputs are not public, so a public re-run is not offered.
Failover, by family
The router tries each route in this order and skips one that is down, out of balance or failing. Steps are shown by family; providers and exact ids are not shown (show settings).
gpt-oss: 3 routes, this alias's own weightsgpt-oss: 1 route, our own endpoint, when one is registeredglm-flash: 1 route, same weights as gaents/glm-flashclaude-haiku: 1 route, same weights as gaents/claude-haiku
From the router's routes table (read 2026-11-19T04:27:48Z). Whether a route is live now is the capacity count above.