agents benchmark
DSBench-Hard Leaderboard
DSBench-Hard is DeepSeek's internal test set of difficult coding-agent problems.
Updated Sep 5, 2026
Models3
Model coverage3
MetricScore
EvidenceB
DSBench-Hard Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelDE | Score67.20% | Percentile100.00% | Participants3 | EvidenceC | Evaluated |
| Rank02 | ModelDE | Score63.60% | Percentile50.00% | Participants3 | EvidenceC | Evaluated |
| Rank03 | ModelDE | Score59.60% | Percentile0.00% | Participants3 | EvidenceC | Evaluated |
DSBench-Hard Highlights
The leading models and scores on this benchmark.
DSBench-Hard Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for DSBench-Hard
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What is DSBench-Hard?
What DSBench-Hard measures and how its scores work.
DSBench-Hard is a benchmark in the agents category consisting of DeepSeek's internal test set of difficult coding-agent problems.
It measures Score, reported as a ratio, on difficult coding-agent problems.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- DSBench-Hard
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about DSBench-Hard.
Which model scores highest on DSBench-Hard?
DeepSeek-V4-Pro-0813 is currently ranked first with 67.20%.
What are the top three models on DSBench-Hard?
The current leaders are DeepSeek-V4-Pro-0813 (67.20%), DeepSeek-V4-Flash-Vision-Exp (63.60%), and DeepSeek-V4-Flash-0731 (59.60%).
Which DSBench-Hard model has the lowest official input price?
DeepSeek-V4-Flash-Vision-Exp has the lowest matched official input price at $0.14 input / $0.28 output per 1M tokens.
Which models are fastest among DSBench-Hard results?
The fastest matched records are DeepSeek-V4-Pro-0813 (45.48 tok/s via DeepSeek), DeepSeek-V4-Flash-Vision-Exp (15.91 tok/s via DeepInfra), and DeepSeek-V4-Flash-0731 (15.91 tok/s via DeepInfra).
Does the highest score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.