NVIDIA Nemotron 3.5 Super Report - BrowseComp (OpenCode): leaderboard
NVIDIA Nemotron 3.5 technical report: BrowseComp run in an OpenCode-based agent harness rather than NVIDIA's NeMo Gym search harness (Exa search, discard-all context) used for its main BrowseComp row. NVIDIA ran all seven models under identical serving settings; score (%) is its own measurement, not a copied peer result.
Metric: Score (%). Source: research.nvidia.com. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 Flash | 81.91 | #77 |
| 2 | Qwen3.8-Flash-Next | 76.07 | #76 |
| 3 | Nemotron 3.5 Super VL (BF16) | 72.35 | #160 |
| 4 | DeepSeek V4 Flash (0731) | 64.77 | #84 |
| 5 | Nemotron 3 Ultra | 61.06 | #245 |
| 6 | Inkling Small | 55.37 | #188 |
| 7 | Qwen 3.5 122B A10B | 45.42 | #184 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nvidia-nemotron-3-5-super-report-browsecomp-opencode · How It Works · Data refreshed daily, snapshot 2026-11-05.