Synth Evals · Development
CyberneticsBench v2
Game and trading-card policies · 10-minute coding sessions.
33Tasks
9Configurations
GPT-6 Luna
Leaderboard
Notes
Scores average completed tasks. Missing scores are excluded; valid zeros are retained. Native task scores are not a calibrated overall rating.
Coverage can differ by effort. Frontier lines require matching evaluation identities and scored-task sets.
Costs cover scored tasks only. Hollow markers indicate partial reports. Unresolved charges and infrastructure are excluded.
| Model | Effort | Coverage | Score | Cost |
|---|---|---|---|---|
| XHigh | 31 / 33 | 60.1% | $1.624 | |
| High | 30 / 33 | 55.1% | $1.503 | |
| Low | 27 / 33 | 45.4% | $0.711 | |
| Medium | 20 / 33 | 45.2% | $0.783 | |
| High | 31 / 33 | 45.1% | $1.309* | |
| XHigh | 23 / 33 | 38.4% | $1.318* | |
| Max | 21 / 33 | 34.4% | $1.588* | |
| Medium | 31 / 33 | 34.0% | $1.219* | |
| Low | 31 / 33 | 32.9% | $0.366* |
Cost vs. Performance
Model
GPT-6 Luna · Low
Score
45.4%
Coverage
27 / 33
Model cost
$0.711
Task scores
33 tasksScore 0–100%
Tasks
Rows 1–8 of 33Filters
| Task | Outcome | Score | Cost |
|---|---|---|---|
| Timeout | — | — | |
| Scored | 27.8% | $0.015 | |
| Scored | 5.0% | $0.014 | |
| Scored | 11.0% | $0.027 | |
| Timeout | — | — | |
| Scored | 72.5% | $0.018 | |
| Scored | 66.3% | $0.068 | |
| Scored | 25.0% | $0.013 |