Includes 694 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 4 rows, muse-glimmer-30b leads 3.
| step-5-preview | Metric | muse-glimmer-30b |
|---|---|---|
| stepfun Unstable | Status | Serving meta |
| 45 | Score | ●64 |
| 50.0% | Uptime | ●91.7% |
| 7.49 s● | First token (median) | 33.4 s |
| 25.4 s● | Full reply (median) | 61.1 s |
| 7.40 s● | Best reply | 26.5 s |
| 10.7 tok/s● | Throughput (median) | 8.2 tok/s |
| 0 | Round wins | 0 |
| #10 | Leaderboard rank | ●#9 |
| 66.7% | Faster when both answered (6 rounds) | 33.3% |
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.
ABBB·ABBBAAB
Astep-5-preview Bmuse-glimmer-30b ·neither answered