Comparing AI models' ability to assemble experimentally verified constructs. Benchmarked against verified Addgene plasmids as the 100% accuracy control.
8
Models Tested
500
Plasmids per Model
100%
Addgene Verified
Compare accuracy, speed, and cost across AI models
Provider
Min Accuracy
AnyChart Axes
Y-Axis
X-Axis
Legend
Ranked by assembly accuracy across 500 verified Addgene plasmids
| # | Model | Accuracy | Speed | Cost | Errors | Complexity |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 4Anthropic | 97.8% | 45s | $0.12 | 3 | |
| 2 | GPT-5OpenAI | 96.5% | 52s | $0.15 | 5 | |
| 3 | Gemini 3 ProGoogle | 95.8% | 38s | $0.08 | 6 | |
| 4 | Claude Sonnet 4Anthropic | 94.2% | 28s | $0.04 | 8 | |
| 5 | DeepSeek V3DeepSeek | 92.1% | 42s | $0.01 | 11 | |
| 6 | GPT-4oOpenAI | 91.3% | 35s | $0.05 | 12 | |
| 7 | Llama 4 405BMeta | 90.8% | 55s | $0.03 | 13 | |
| 8 | Gemini 3 FlashGoogle | 89.5% | 18s | $0.02 | 15 |
Benchmarks conducted using verified Addgene plasmid sequences as ground truth.
Learn about our methodology