Nemotron-4 340B Instruct: benchmark results
NVIDIA's Nemotron-4 340B Instruct, fine-tuned from Nemotron-4-340B-Base via SFT, DPO and RPO alignment on mostly synthetic data, for English chat and synthetic-data-generation pipelines with a 4,096-token context. Provider: NVIDIA. Released 2024-06-13. Access: Open.
Aggregate Rating 1542 ± 20, rank #632 of 1635 rated models, from 44 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| VNTL Leaderboard | 72.79 | Accuracy (%) | 93 |
| pfgen-bench - Completion Mode - Helpfulness | 0.43 | Helpfulness Score | 92.2 |
| pfgen-bench - Completion Mode - Score | 0.72 | pfgen Score (mean of three) | 92 |
| pfgen-bench - Completion Mode - Truthfulness | 0.91 | Truthfulness Score | 91.4 |
| pfgen-bench - Completion Mode - Fluency | 0.82 | Fluency Score | 90.7 |
| pfgen-bench - QA Mode - Helpfulness | 0.44 | Helpfulness Score | 85.5 |
| pfgen-bench - QA Mode - Score | 0.7 | pfgen Score (mean of three) | 83.6 |
| pfgen-bench - QA Mode - Fluency | 0.77 | Fluency Score | 82.4 |
| pfgen-bench - QA Mode - Truthfulness | 0.91 | Truthfulness Score | 81.5 |
| WildBench | 47.67 | WB Score Task-Macro | 71 |
| BenchTable - Reasoning | 39.7 | Weighted Score (%) | 49 |
| Wolfram LLM Benchmarking Project | 35.6 | Correct Functionality (%) | 37.8 |
Interactive version: theaggregate.ai/model?slug=nemotron-4-340b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-15.