DSCodeBench
Models write Python functions for 1,000 data science problems covering ten libraries. The evaluator checks generated code against reference outputs.
The independent HERO team evaluated 482 held-out problems with the official comparator and 200 seeded inputs per problem.
- Made by
- Shuyin Ouyang and collaborators
- Released
- May 2025
- Kind
- Data science and SQL
- Level
- University
- Scored
- Checked automatically
- Answers
- Code
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use, Generated
- Refresh
- One fixed set
- Run by
- Anyone
- Last model added
- November 2024
- Scores refreshed
- November 20th, 2026
GPT-4o (snapshot unspecified) tops the leaderboard with 39.2% Qwen2.5-Coder-32B-Instruct leads with 22.9% among models released by November 2024
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Qwen2.5-Coder-32B-Instruct, at 22.9%. This view shows only models we track, each at its best.
The newest model with a score here came out in November 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 39.2 GPT-4o (snapshot unspecified)
- 34.2 GPT-4o-mini (snapshot unspecified)
- 30.7 GPT-3.5-turbo (snapshot unspecified)
- 22.9
Qwen2.5-Coder-32B-Instruct
- 22.2 DeepSeek-Coder-33B-Instruct
- 21.3 Qwen2.5-Coder-14B-Instruct
- 20.5 DeepSeek-Coder-V2-Lite-Instruct
- 16.3 DeepSeek-Coder-6.7B-Instruct
- 11.6 Qwen2.5-Coder-7B-Instruct
- 7.60 DeepSeek-Coder-1.3B-Instruct
- 1 GPT-4o (snapshot unspecified) Temperature 0.2 39.2%
- 2 GPT-4o-mini (snapshot unspecified) Temperature 0.2 34.2%
- 3 GPT-3.5-turbo (snapshot unspecified) Temperature 0.2 30.7%
- 4
Qwen2.5-Coder-32B-Instruct Temperature 0.2 22.9%
- 5 DeepSeek-Coder-33B-Instruct Temperature 0.2 22.2%
- 6 Qwen2.5-Coder-14B-Instruct Temperature 0.2 21.3%
- 7 DeepSeek-Coder-V2-Lite-Instruct Temperature 0.2 20.5%
- 8 DeepSeek-Coder-6.7B-Instruct Temperature 0.2 16.3%
- 9 Qwen2.5-Coder-7B-Instruct Temperature 0.2 11.6%
- 10 DeepSeek-Coder-1.3B-Instruct Temperature 0.2 7.60%
Source: The DSCodeBench paper
Source: The DSCodeBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o (snapshot unspecified) | 39.2% | Temperature 0.2 | Official leaderboard | |
| 2 | GPT-4o-mini (snapshot unspecified) | 34.2% | Temperature 0.2 | Official leaderboard | |
| 3 | GPT-3.5-turbo (snapshot unspecified) | 30.7% | Temperature 0.2 | Official leaderboard | |
| 4 | | 22.9% | Temperature 0.2 | Nov 2024 | Official leaderboard |
| 5 | DeepSeek-Coder-33B-Instruct | 22.2% | Temperature 0.2 | Official leaderboard | |
| 6 | Qwen2.5-Coder-14B-Instruct | 21.3% | Temperature 0.2 | Nov 2024 | Official leaderboard |
| 7 | DeepSeek-Coder-V2-Lite-Instruct | 20.5% | Temperature 0.2 | Jun 2024 | Official leaderboard |
| 8 | DeepSeek-Coder-6.7B-Instruct | 16.3% | Temperature 0.2 | Official leaderboard | |
| 9 | Qwen2.5-Coder-7B-Instruct | 11.6% | Temperature 0.2 | Official leaderboard | |
| 10 | DeepSeek-Coder-1.3B-Instruct | 7.60% | Temperature 0.2 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 22.9% | Temperature 0.2 | Nov 2024 | Official leaderboard |
| 2 | Qwen2.5-Coder-14B-Instruct | 21.3% | Temperature 0.2 | Nov 2024 | Official leaderboard |
| 3 | DeepSeek-Coder-V2-Lite-Instruct | 20.5% | Temperature 0.2 | Jun 2024 | Official leaderboard |
Rankings by part
We read 10 parts from the benchmark’s paper, each ranking models on one part of the test, such as NumPy, Pandas and Scipy. Qwen2.5-Coder-14B-Instruct leads 5 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
NumPy
3 models we track have a score here.
- 1
Qwen2.5-Coder-32B-Instruct 33.6%
- 2 Qwen2.5-Coder-14B-Instruct 27.5%
- 3 DeepSeek-Coder-V2-Lite-Instruct 19.6%
Pandas
3 models we track have a score here.
- 1 Qwen2.5-Coder-14B-Instruct ↑1 19.2%
- 2
Qwen2.5-Coder-32B-Instruct ↓1 17.8%
- 3 DeepSeek-Coder-V2-Lite-Instruct 14.9%
Scipy
3 models we track have a score here.
- 1 Qwen2.5-Coder-14B-Instruct ↑1 3.60%
- 2
Qwen2.5-Coder-32B-Instruct ↓1 3.60%
- 3 DeepSeek-Coder-V2-Lite-Instruct 2.70%
Scikit-learn
3 models we track have a score here.
- 1 DeepSeek-Coder-V2-Lite-Instruct ↑2 40.1%
- 2 Qwen2.5-Coder-14B-Instruct 32.7%
- 3
Qwen2.5-Coder-32B-Instruct ↓2 29.9%
Matplotlib
3 models we track have a score here.
- 1 Qwen2.5-Coder-14B-Instruct ↑1 2.90%
- 2 DeepSeek-Coder-V2-Lite-Instruct ↑1 2.20%
- 3
Qwen2.5-Coder-32B-Instruct ↓2 2.20%
Seaborn
3 models we track have a score here.
- 1 Qwen2.5-Coder-14B-Instruct ↑1 17.7%
- 2
Qwen2.5-Coder-32B-Instruct ↓1 14.9%
- 3 DeepSeek-Coder-V2-Lite-Instruct 12.9%
TensorFlow
3 models we track have a score here.
- 1
Qwen2.5-Coder-32B-Instruct 43%
- 2 Qwen2.5-Coder-14B-Instruct 36.7%
- 3 DeepSeek-Coder-V2-Lite-Instruct 33.3%
PyTorch
3 models we track have a score here.
- 1 DeepSeek-Coder-V2-Lite-Instruct ↑2 46.5%
- 2
Qwen2.5-Coder-32B-Instruct ↓1 44.9%
- 3 Qwen2.5-Coder-14B-Instruct ↓1 39.6%
Keras
3 models we track have a score here.
- 1 Qwen2.5-Coder-14B-Instruct ↑1 19.6%
- 2 DeepSeek-Coder-V2-Lite-Instruct ↑1 17.9%
- 3
Qwen2.5-Coder-32B-Instruct ↓2 17.3%
LightGBM
3 models we track have a score here.
- 1
Qwen2.5-Coder-32B-Instruct 14.2%
- 2 DeepSeek-Coder-V2-Lite-Instruct ↑1 11.1%
- 3 Qwen2.5-Coder-14B-Instruct ↓1 8.60%
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The DSCodeBench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Temperature 0.2 | 22.9% | ||||
| Qwen2.5-Coder-14B-Instruct | Temperature 0.2 | 21.3% | |||
| DeepSeek-Coder-V2-Lite-Instruct | Temperature 0.2 | 20.5% | |||
| DeepSeek-Coder-V2-Lite-Instruct | Temperature 0.6 | 19.6% | |||
| Temperature 0.6 | 15.9% | ||||
| Qwen2.5-Coder-14B-Instruct | Temperature 0.6 | 5.50% |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The DSCodeBench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Temperature 0.2 | 22.9% | ||||
| Qwen2.5-Coder-14B-Instruct | Temperature 0.2 | 21.3% | |||
| DeepSeek-Coder-V2-Lite-Instruct | Temperature 0.2 | 20.5% | |||
| DeepSeek-Coder-V2-Lite-Instruct | Temperature 0.6 | 19.6% | |||
| Temperature 0.6 | 15.9% | ||||
| Qwen2.5-Coder-14B-Instruct | Temperature 0.6 | 5.50% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The DSCodeBench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official DSCodeBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Coding
- SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
- LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests.
DeepSeek-V4-Pro 93.5%
- SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
- DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
- SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases.
Seed2.1 Pro 59.8%
- Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance.
DeepSeek V4.1 Flash 3,471
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in November 2024, for Qwen2.5-Coder-32B-Instruct. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (10 entries).
- Pass Models scored 3 models we track, plus 7 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in November 2024, for Qwen2.5-Coder-32B-Instruct.
- Note Frontier models None of its models are on the capabilities index.
- Pass Headroom The best is 22.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track and 7 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.