Coding · Benchmark profile
A set of 164 handwritten Python function-generation problems. HumanEval is useful as a historical floor check, but BenchLM's current exact-source table is too small to support a broad frontier-coding verdict.
Data verified 21 Jul 2026 · Methodology 1.6.0
Benchmark score on Evaluating Large Language Models Trained on Code
DeepSeek V4 Pro Base leads at 76.8%, followed by Soofi S 30B-A3B (73.8%) and DeepSeek V4 Flash Base (69.5%).
Visual analysis
Switch between model placement, score distribution and descriptive provider averages. Every view uses the same sourced leaderboard.
3 ranked models · higher is better
| Model | Rank | Provider | Score |
|---|---|---|---|
| DeepSeek V4 Pro Base | #1 | DeepSeek | 76.8% |
| Soofi S 30B-A3B | #2 | Soofi Project | 73.8% |
| DeepSeek V4 Flash Base | #3 | DeepSeek | 69.5% |
One best score per model · higher is better
| Rank | Model | Provider | License | Evidence use | Score |
|---|---|---|---|---|---|
| #1 | DeepSeek V4 Pro Base deepseek-v4-pro-base | DeepSeek | open | Estimated reference | 76.8% |
| #2 | Soofi S 30B-A3B soofi-s-30b-a3b | Soofi Project | open | Estimated reference | 73.8% |
| #3 | DeepSeek V4 Flash Base deepseek-v4-flash-base | DeepSeek | open | Estimated reference | 69.5% |
The top of this snapshot is led by DeepSeek V4 Pro Base at 76.8%; third place is 7.3 points behind. The top-3 spread is 7.3 points.
About Evaluating Large Language Models Trained on Code
A set of 164 handwritten Python function-generation problems. HumanEval is useful as a historical floor check, but BenchLM's current exact-source table is too small to support a broad frontier-coding verdict. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.
Open benchmark source ↗FAQ
A set of 164 handwritten Python function-generation problems. HumanEval is useful as a historical floor check, but BenchLM's current exact-source table is too small to support a broad frontier-coding verdict.
DeepSeek V4 Pro Base by DeepSeek currently leads with 76.8%.
3 models in the LuminaBench cohort have a qualifying score on this benchmark.
Related
