Llama-3.3 Nemotron Super 49B v1
Llama-3.3-Nemotron-Super-49B-v1 is a large language model (LLM) derived from Meta Llama-3.3-70B-Instruct. It's post-trained for reasoning, chat, RAG, and tool calling, offering a balance between accuracy and efficiency (optimized for single H100). It underwent multi-phase post-training including SFT and RL (RLOO, RPO).
Benchmark results
| Benchmark | Score | Tags | Source |
|---|---|---|---|
| AIME 2025 | 58.4% | self-reported llm-stats | link → |
| Arena Hard | 88.3% | self-reported llm-stats | link → |
| BFCL v2 | 73.7% | self-reported llm-stats | link → |
| GPQA | 66.7% | self-reported llm-stats | link → |
| MATH-500 | 96.6% | self-reported llm-stats | link → |
| MBPP | 91.3% | self-reported llm-stats | link → |
| MT-Bench | 0.917 | self-reported llm-stats | link → |