Source
Server-written Reelipal generation audit log. Each terminal request is counted once; settlement-only repair events are excluded.
What completed successfully in real production?
A privacy-safe reliability benchmark built from Reelipal’s server-written audit log—not a hand-picked demo set.
June 25, 2026 — July 13, 2026
251Terminal requests
97.2%Completion rate
244Succeeded
7Failed
4,745Credits processed
Data
Download machine-readable data ↓Model-level results
Model rows require at least 10 attempts. Eleven lower-volume model/workflow groups remain in the overall total but are not ranked separately.
| Workflow | Model | Attempts | Succeeded | Failed | Completion |
|---|---|---|---|---|---|
| image | Nano Banana Pro | 53 | 51 | 2 | 96.2% |
| image | Nano Banana 2 | 52 | 52 | 0 | 100% |
| video | Veo 3.1 Fast | 27 | 27 | 0 | 100% |
| video | Kling 2.5 Turbo | 21 | 19 | 2 | 90.5% |
| lipsync | Sync LipSync v3 | 18 | 18 | 0 | 100% |
| video | Seedance 2.0 | 17 | 17 | 0 | 100% |
| video | Hailuo 02 | 10 | 10 | 0 | 100% |
| voice | MiniMax Speech 2.8 HD | 10 | 10 | 0 | 100% |
Methodology
Every included terminal request is counted. Models appear separately only after reaching the published minimum sample.
Privacy
No user ids, prompts, creation ids, media URLs or error text are selected for this report.
What this does not measure
Success means a request completed and returned deliverable media. It does not score visual quality or prompt adherence.