Includes 693 hourly rounds up to 2026-11-16 from NIMStats by Mauro Druwel (source), measured with a different prompt and timeout. Our own probe adds a round every 30 minutes.
Model comparison
Head-to-head from live probes, last 24 hours.
Head-to-head stats
step-5-preview leads 7 rows, gpt-oss-20b leads 0.
| step-5-preview | Metric | gpt-oss-20b |
|---|---|---|
| stepfun Unstable | Status | Down openai |
| 43● | Score | 0 |
| 46.2%● | Uptime | 0% |
| 7.49 s● | First token (median) | n/a |
| 25.4 s● | Full reply (median) | n/a |
| 7.40 s● | Best reply | n/a |
| 10.7 tok/s● | Throughput (median) | n/a |
| 0 | Round wins | 0 |
| #10● | Leaderboard rank | #14 |
| n/a | Faster when both answered (0 rounds) | n/a |
marks the better value in each row
Full reply time overlay
Each probe, both models on one axis. Lower is better.
Win timeline
Per round, oldest left: which model replied faster.
A·A··AA··AA··
Astep-5-preview Bgpt-oss-20b ·neither answered