Nemotron-3.5 Bolt hits 9B active params, within 1% of Super on agentic benchmark
ryanshrout · x · 2026-10-27
Signal65 benchmarked NVIDIA's new Nemotron-3.5 Bolt 74B-A9B on PINNACLE: with 9B active parameters (vs Super's 12B) it scored 927, within 1% of Nemotron-3.5 Super's 937, finishing 52.5% of multi-step enterprise jobs (vs 50.7%) and 93.5% of the 128K retrieval corpus, with 46% fewer weighted errors than Nemotron-3.5 Lightning and the best token efficiency in the family (43,821 output tokens per correct task). Nemotron-3-Ultra remains top at 1,045. PINNACLE uses procedurally generated sandboxes and answer keys with live agentic sessions; NVIDIA and AMD executives endorsed the methodology.
More from Models
- Steering vectors turn Qwen3.5-4B into "Golden Gate Qwen" with a known recipe — burny_tech · 2026-10-27
- Experiment finds 9 'gender neurons' in Qwen3.5-4B that flip its self-identity — burny_tech · 2026-10-27
- MiniMax M3.1 Flash lands on Together AI with LoRA post-training for SFT and RL — MiniMax_AI · 2026-10-27
- Ethan Mollick resurfaces Copilot's classic strawberry fail: 'Two and a half years from this' — aran_nayebi · 2026-10-27
- Claude Opus 5.5 generates explainer video for OpenAI's three discoveries — iScienceLuvr · 2026-10-27
- X forces "Paid Partnership" label on unpaid Kimi K3.1 Pro tweet, locks author's account — RazRazcle · 2026-10-27