NVIDIA Nemotron 3.5 Super and Lightning technical report

79-page vendor technical report, fetched 2026-10-18. Defines Nemotron 3.5 Super as a 120B-total / 12B-active hybrid Mamba-Transformer initialized from Nemotron 3 Super, and Lightning as 30B-total / 3B-active. Table 10 is the evaluation suite: OpenCode harness for SWE-bench, Terminus 2 with a 3 hour task timeout for Terminal-Bench 2.1. N3.5-Super column: SWE-bench Verified 71.10, SWE-bench Pro 51.78, Terminal-Bench 2.1 60.67, GPQA Diamond 89.52. Same table, Qwen3.5 122B-A10B column: SWE-bench Verified 71.87, SWE-bench Pro 45.60, Terminal-Bench 2.1 48.70, GPQA Diamond 86.74. Qwen3.8 Flash-Next column is higher on all four (72.60, 60.37, 85.80, 92.68). Section 4.2 restates 512 routed experts per MoE layer but does not state how many are activated per token. All numbers are vendor measurements, not an independent re-run.