Stardust Flash-01
Results & cost
Performance per dollar on 60 held-out tasks.
One full pass over all 60 held-out tasks per model. Spend is estimated from each provider's list pricing.
| Model | Solved | Pass@1 | Cost of a full run |
|---|---|---|---|
| Claude Opus 5 High (reference) | 43/60 | 71.7% | $74.68 |
| Stardust Flash-01 | 40/60 | 66.7% | $2.38 |
| GLM-5.3-Flash (base model) | 36/60 | 60.0% | $4.56 |
On our internal held-out evaluation Flash-01 beats its base model by a steady 5–8%, and reaches about 93% of Claude Opus 5 High's score at about 3% of the cost.
Why our own benchmark#
Public cyber benchmarks tend to reduce security work to tidy puzzles. Our environments resemble the engagements we run for customers, down to real exploitable bugs we found, reported responsibly and in many cases were paid bounties for.