ResearchBenchmarksGrade School Math Reasoning on GSM8K No CoT augmented (test)Follow0.2482AccuracyTurboConn0.0649520.1125260.16010.207674Feb 20, 2026Evaluation ResultsMethodMethodLinksAccuracyTurboConnModel=Llama 3.1 8BModel=Llama 3.1 8B2026.020.2482BaselineModel=Llama 3.1 8BModel=Llama 3.1 8B2026.020.2392TurboConnModel=Qwen 3 1.7BModel=Qwen 3 1.7B2026.020.2031BaselineModel=Qwen 3 1.7BModel=Qwen 3 1.7B2026.020.159TurboConnModel=Llama 3.2 1BModel=Llama 3.2 1B2026.020.0832BaselineModel=Llama 3.2 1BModel=Llama 3.2 1B2026.020.072