Across 13 shared benchmarks, Claude Sonnet 4.5 scores higher on 3 and nvidia-nemotron-3-ultra-550b-a55b on 10. The widest gap is browsecomp, where nvidia-nemotron-3-ultra-550b-a55b scores 44.4 against 24.1. nvidia-nemotron-3-ultra-550b-a55b is the cheaper of the two on tracked API pricing ($0.60 against $3.00 per million input tokens).
| Benchmark | Claude Sonnet 4.5 | nvidia-nemotron-3-ultra-550b-a55b |
|---|---|---|
| browsecomp | 24.1 | 44.4 |
| gdpval | 40.4 | 46.7 |
| GPQA Diamond | 83.4 | 87 |
| HLE | 19.8 | 26.7 |
| HLE (with tools) | 32 | 37.4 |
| imo_answer_bench | 65.9 | 92.3 |
| LiveCodeBench v6 | 64 | 89 |
| longbench_v2 | 61.8 | 61.9 |
| MMLU-Pro | 88.2 | 86.8 |
| multichallenge | 55.3 | 63.8 |
| SWE-bench Multilingual | 68 | 67.7 |
| SWE-bench Verified | 82 | 70.7 |
| Terminal-Bench 2.1 | 55.8 | 56.4 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (anthropic-official, direct), otherwise the lowest tracked offer.