DeepSeek DSBench Hard (DSBench-Hard)
We show this table for reference; we do not rank on it.
DeepSeek's internal hard coding-agent benchmark.
Benchmark score on DSBench-Hard — September 27, 2026
We compile the DSBench-Hard rows from provider self-reports. DeepSeek V4 Pro 0813 leads the table at 67.2%, followed by DeepSeek V4 Flash 0731 (59.6%). We do not use these results to rank models overall.
DeepSeek V4 Pro 0813
DeepSeek
DeepSeek V4 Flash 0731
DeepSeek
2 modelsCodingCurrentDisplay onlyUpdated September 27, 2026
Benchmark score table (2 models)
ScoreAbout DSBench-Hard
Year
2026
Tasks
Internal hard coding-agent tasks
Format
Provider-reported score
Difficulty
Advanced coding-agent challenges
DeepSeek labels DSBench-Hard as an internal benchmark. BenchLM stores the provider-reported exact result as display-only launch evidence and does not compare it with public benchmark leaderboards.
Freshness and provenance
Version
DSBench-Hard 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does DSBench-Hard measure?
DeepSeek's internal hard coding-agent benchmark.
Which model scores highest on DSBench-Hard?
DeepSeek V4 Pro 0813 by DeepSeek currently leads with a score of 67.2% on DSBench-Hard.
How many models are evaluated on DSBench-Hard?
2 AI models have been evaluated on DSBench-Hard on BenchLM.
Compare top models on DSBench-Hard
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 2,000+ readers.
One email each week. Unsubscribe anytime.