A new open-weight model, Laguna S 2.1, with 118 billion parameters, scored 70.2% on Terminal-Bench 2.1, outperforming larger models like DeepSeek-V4-Pro-Max (1.6 trillion parameters, 64.0%), Inkling (975B, 63.8%), and Nemotron 3 Ultra (550B, 56.4%). On the harder DeepSWE benchmark, Laguna S 2.1 scored 40.4 against DeepSeek-V4-Pro-Max’s 9.0, representing a 13x parameter deficit paired with a 4x score advantage. To address potential skepticism about the results, Poolside published every trajectory from every trial in the final evaluation set.