DSBench-Hard

Name: DSBench-Hard Leaderboard — AI Model Scores
Creator: LLM Stats
License: https://llm-stats.com/legal/terms-of-service

Progress Over Time

Interactive timeline showing model performance evolution on DSBench-Hard

State-of-the-art frontier

Open

Proprietary

DSBench-Hard Leaderboard

3 models

			Context	Cost
1	DeepSeek-V4-Pro-0813 DeepSeek	1.6T	1.0M	$0.43 / $0.87
2	DeepSeek-V4-Flash-Vision-ExpNew DeepSeek	—	1.0M	$0.22 / $0.66
3	DeepSeek-V4-Flash-0731 DeepSeek	304B	1.0M	$0.09 / $0.18

Notice missing or incorrect data?

About this benchmark

What is DSBench-Hard?

DSBench-Hard is DeepSeek's internal test set of difficult coding-agent problems.

DSBench-Hard is a text benchmark evaluating models on agents and code tasks. LLM Stats tracks 3 models on this benchmark, scored on a 0–1 scale. The current average is 0.6, with the leader at 0.7.

Compare leaders on the best AI for agents and best AI for code leaderboards.

Current leaders

DeepSeek-V4-Pro-0813 from DeepSeek currently leads the DSBench-Hard leaderboard with a score of 0.672 across 3 evaluated AI models.

DeepSeek-V4-Pro-0813DeepSeek67.2%

DeepSeek-V4-Flash-Vision-ExpDeepSeek63.6%

DeepSeek-V4-Flash-0731DeepSeek59.6%

FAQ

Common questions about the DSBench-Hard benchmark and leaderboard.

What is the DSBench-Hard benchmark?

DSBench-Hard is DeepSeek's internal test set of difficult coding-agent problems.

What is the DSBench-Hard leaderboard?

The DSBench-Hard leaderboard ranks 3 AI models based on their performance on this benchmark. Currently, DeepSeek-V4-Pro-0813 by DeepSeek leads with a score of 0.672. The average score across all models is 0.635.

What is the highest DSBench-Hard score?

The highest DSBench-Hard score is 0.672, achieved by DeepSeek-V4-Pro-0813 from DeepSeek.

How many models are evaluated on DSBench-Hard?

3 models have been evaluated on the DSBench-Hard benchmark, with 0 verified results and 3 self-reported results.

What categories does DSBench-Hard cover?

DSBench-Hard is categorized under agents and code. The benchmark evaluates text models.

What is the best open-source model on DSBench-Hard?

DeepSeek-V4-Pro-0813 by DeepSeek is the top-ranked open-source model on DSBench-Hard, with a score of 0.672 (rank #1).

Which model offers the best value on DSBench-Hard?

Among models scoring within 10% of the leader, DeepSeek-V4-Flash-Vision-Exp from DeepSeek is the cheapest, at $0.22 per million input tokens with a score of 0.636.

How recent are the DSBench-Hard leaderboard results?

The DSBench-Hard leaderboard was last updated in August 2026 and currently includes 3 evaluated models.