Replicate · Predictions dashboard
Replicate made running open models one line of code. This rebuild keeps the predictions list exactly as it is — and gives it the same clarity as the CLI.
| Model | Status | Duration | Hardware | Source | Created | ID |
|---|---|---|---|---|---|---|
| black-forest-labs/flux-1.1-pro | Succeeded | 8.2s | A100 80GB | API | 09:41:50 | xk2p9qerhn |
| meta/llama-3-70b-instruct | Succeeded | 12.4s | A100 80GB | API | 09:41:12 | d4m1vz0wqr |
| stability-ai/sdxl | Failed | 2.1s | A40 | API | 09:40:38 | p8c3nx5jte |
| openai/whisper | Succeeded | 34.8s | T4 | Web | 09:39:47 | w7f2kd9mbs |
| black-forest-labs/flux-schnell | Processing | — | H100 | API | 09:39:21 | r5h8qa2vcn |
| meta/llama-3-70b-instruct | Succeeded | 9.7s | A100 80GB | API | 09:38:02 | t9j4ye6zlp |
| mistralai/mixtral-8x7b-instruct-v0.1 | Canceled | 1.0s | CPU | Web | 09:36:44 | g2b7mw4xdf |
| stability-ai/sdxl | Succeeded | 6.6s | A40 | API | 09:35:19 | k6n1rc8uah |
| openai/whisper | Failed | 0.4s | T4 | API | 09:33:58 | z3v5ps7qek |
Rebuilt by Kyre — same layout, same data, same information architecture. Only the execution changed.
What changed — and why it matters
You open this screen to answer two questions: what died, and what's slow. I moved duration next to status so both are answered in one glance — column order is a decision about how people actually read, not alphabetical habit.
A100 vs T4 is the difference in your invoice. The chips make hardware scannable so you can spot the job burning an H100 on work a T4 could do. An engineer renders the field; deciding it deserves visual weight is the outside eye.
You scan for flux-1.1-pro, not black-forest-labs. Dimming the owner is a one-line CSS change and a real decision about where your eye lands in a list of 748 — that ranking of attention is what you're paying for.