DSBench-Hard †Browse 284 families
DSBench-Hard †
Can the model complete substantial programming work under this benchmark's agent setup?
Version / measure
- Domain
- Coding
- Ranked models
- 5
- Reported values
- 5
- Reports
- 1 · lab
- Direction
- Higher is better
- Status
- Version not specified
Rankings
Best reported score per model across effort settings
| # | Model | Best setting | Score |
|---|---|---|---|
| 1 | Claude Opus 4.8Anthropic | reportedcost n/a1 value · Jul 31 · source | 71.7% |
| 2 | DeepSeek V4 Flash 0731DeepSeekOW | maxcost n/a1 value · Jul 31 · source | 59.6% |
| 3 | GLM-5.2Z.aiOW | reportedcost n/a1 value · Jul 31 · source | 54.5% |
| 4 | DeepSeek V4 ProDeepSeekOW | reportedcost n/a1 value · Jul 31 · source | 31.1% |
| 5 | DeepSeek-V4-FlashDeepSeekOW | reportedcost n/a1 value · Jul 31 · source | 25.8% |
Effort curve
Score vs. reported cost. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundaryinternal methodology
Can the model complete substantial programming work under this benchmark's agent setup?
Software implementation, debugging, or repository work under the published evaluation protocol. Higher is better. The value is the percentage reported in this lab's table.
This is a publisher-defined internal evaluation. The task set or grading details are not fully public, so treat it as directional evidence.
† DeepSeek's internal difficult coding-agent test set; the source retains the dagger marker.
Method / source ↗