VoidGPT
VoidGPT Leaderboard
Evaluation results across VoidGPT models — sortable, searchable, and always up to date.
- Models
- —
- Best average
- —
- Updated
- —
LLM Benchmark
Compare evaluation results across models.
About
Methodology and reproducibility information.
How it works
Each model is evaluated on a fixed set of benchmarks. The table shows one score per benchmark, and Average is the mean of the benchmark scores.
Benchmarks
MBPP+
Mostly Basic Python Problems Plus — an enhanced version of the MBPP code-generation benchmark with sanitized problems and ~35× more test cases per problem. The model writes a Python function that is verified by unit tests. Higher means more reliable code generation.
GSM8K
Grade School Math 8K (OpenAI) — 8.5K grade-school math word problems requiring 2–8 steps of elementary arithmetic. The test set has 1,319 problems. Higher means better multi-step math reasoning.
GSM8K-Platinum
A revised version of the full GSM8K test set (Madry Lab): frontier-model disagreements were manually reviewed — 110 poorly written questions removed, 10 mislabeled answers corrected — yielding a 1,209-question drop-in replacement with less label noise. Higher means better math reasoning on verified data.
Reproducibility
Run the evaluation harness on the exact model revision listed in the table to reproduce a score. Scores are reported in the 0–100 range.
Add a model
Append an entry to static/models.json — one object per model. See the README for the field format.