Quantum API Drift: leaderboard
The quantum-api-drift benchmark (Khalifa University, 2026) asks models to write Qiskit code for a named SDK version and runs it in that version, three samples per prompt at temperature 0.8. The score is Pass@1 in percent averaged over the Qiskit 0.43, 1.3 and 2.0 targets, each with 50 Qiskit HumanEval-derived tasks; higher is better.
Metric: Mean version fidelity (%): Pass@1 on the prompted Qiskit version, averaged over the Qiskit 0.43, 1.3 and 2.0 targets (unbiased estimator from 3 samples per task over 50 Qiskit HumanEval-derived tasks); REST API access at temperature 0.8 with a 1,024-token output cap; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 66.89 | #79 |
| 2 | GPT-5.3 Codex | 60.1 | #116 |
| 3 | Grok 4.20 | 57.78 | #297 |
| 4 | GPT-5.4 | 55.33 | #123 |
| 5 | Claude Sonnet 4.6 | 50.67 | #161 |
| 6 | Kimi K2 | 49.56 | #439 |
| 7 | Nemotron 3 Super | 44.52 | #768 |
| 8 | GPT-OSS-120B | 43 | #700 |
| 9 | Gemma 4 31B (IT) | 34.44 | #342 |
| 10 | Qwen3 Coder | 33.78 | #586 |
| 11 | Devstral 2 | 32.93 | #1003 |
| 12 | GPT-5.4 Nano | 32.89 | #635 |
| 13 | Qwen 3 32B | 28.44 | #960 |
| 14 | GLM-4.7 Flash | 22.89 | #1259 |
| 15 | Llama 3.3 70B Instruct | 17.33 | #1357 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=quantum-api-drift · How It Works · Data refreshed daily, snapshot 2026-11-13.