RealFeedApp QBE raises GPT-6 Astra pass@1 to 94.1%
Researchers tested 15 models across five task suites and the new QBE benchmark, revealing that evaluator design significantly impacts scores. The QiskitHumanEval-Standard model achieved 98.7% pass@1 on standard tasks but only 63.8% on hard variants.
#llm#quantum computing#benchmarking#code generation#evaluation
