80.0%
Enterprise HumanEval
88.0%
Enterprise MBPP
89.9%
Enterprise COBOL
82.9%
V1 HumanEval
Quick Examples
Model Info
Model
Bygheart Coder Enterprise
Parameters
27B
Training Data
CPT + COBOL/enterprise SFT
Context Length
32K tokens
Benchmark Results
Evaluated June 2026 on Modal H100. In-house execution harness.
HumanEval
80.0%Python function generation (164 problems)
Enterprise (27B)
80.0%
Coder V1 (7B)
82.9%
MBPP
88.0%Basic Python problems (257 problems)
Enterprise (27B)
88.0%
Coder V1 (7B)
75.1%
COBOL V2
89.9%Enterprise COBOL benchmark (n=50)
Enterprise (27B CPT+SFT)
89.9%
COBOL V1/V2 (7B)
70.0%
Comparison with Other Models
June 2026 · in-house harness| Model | Size | HumanEval | MBPP | COBOL |
|---|---|---|---|---|
| Bygheart Coder Enterprise ✓ Live | 27B | 80.0% | 88.0% | 89.9% |
| Bygheart Coder V1 (paused) | 7B | 82.9% | 75.1% | — |
| GPT-4o | ~1.8T | 90.2% | 87.8% | — |
| Claude Sonnet 4 | ~175B | 93.5% | 90.1% | — |
| DeepSeek Coder 6.7B | 6.7B | 78.6% | 74.2% | — |
| CodeLlama 34B | 34B | 48.8% | 55.0% | — |