ResearchBenchmarksNLP Tasks on 11 NLP Tasks Symbol-Tuning (held-out)Follow86.4AccuracyICL+FT74.4477.54580.6583.755Dec 22, 2025Evaluation ResultsMethodMethodLinksAccuracyICL+FTModel Size=27BModel Size=27B2025.1286.4ICL+FTModel Size=9BModel Size=9B2025.1285.1ICL+FTModel Size=2BModel Size=2B2025.1284.3FT-OnlyModel Size=27BModel Size=27B2025.1282.6ICL-OnlyModel Size=27BModel Size=27B2025.1282.5ICL-OnlyModel Size=9BModel Size=9B2025.1282.1FT-OnlyModel Size=9BModel Size=9B2025.1281.2FT-OnlyModel Size=2BModel Size=2B2025.1277.7ICL-OnlyModel Size=2BModel Size=2B2025.1274.9