# NVIDIA Nemotron 3 Ultra As of October 21, 2026, NVIDIA Nemotron 3 Ultra ranks #63 of 123 models overall with 53.1 ±15.7 on the LLMBenchmark Index. Nemotron 3 Ultra ranks #26 of 108 in math, its strongest category. NVIDIA Nemotron 3 Ultra costs $0.60 per million input tokens and $2.40 per million output tokens. Nemotron 3 Ultra was released on June 4, 2026. Page: https://llmbenchmark.ai/models/nvidia-nemotron-3-ultra-550b-a55b Data from October 21, 2026. Every figure below names the board it was read from and the day. - Maker: NVIDIA - Released: June 4, 2026 - Price per million input tokens: $0.60 - Price per million output tokens: $2.40 - Context window: 262K tokens - Overall rank: 63 of 123, by the LLMBenchmark Index - LLMBenchmark Index: 53.1 ±15.7 (90% margin, from 5 of 7 inputs) - Artificial Analysis Intelligence Index: 22.9 - Vibe Coding Index: 30.0 (from 3 of 6 coding benchmarks) Benchmark results (70 of 195): - GPQA Diamond: 86.1% (source: GPQA (Vals AI), read October 21, 2026) - MMLU Pro: 85.8% (source: MMLU Pro (Vals AI), read October 21, 2026) - AA Intelligence Index: 22.9 (source: Artificial Analysis (via OpenRouter), read October 21, 2026) - AA Coding Index: 49.3 (source: Artificial Analysis (via OpenRouter), read October 21, 2026) - AA Agentic Index: 20.1 (source: Artificial Analysis (via OpenRouter), read October 21, 2026) - Design Arena: websites: 1145 Elo (source: Design Arena, read October 21, 2026) - Design Arena: UI components: 1139 Elo (source: Design Arena, read October 21, 2026) - Design Arena: game dev: 1138 Elo (source: Design Arena, read October 21, 2026) - Design Arena: data viz: 1141 Elo (source: Design Arena, read October 21, 2026) - Design Arena: 3D: 1132 Elo (source: Design Arena, read October 21, 2026) - Design Arena: all code categories: 1147 Elo (source: Design Arena, read October 21, 2026) - Design Arena: SVG: 1070 Elo (source: Design Arena, read October 21, 2026) - Design Arena: ASCII art: 1100 Elo (source: Design Arena, read October 21, 2026) - LiveCodeBench (Vals AI): 86.0% (source: Vals AI boards, read October 21, 2026) - SWE-bench Verified (Vals AI): 69.0% (source: Vals AI boards, read October 21, 2026) - Vibe Code Bench (Vals AI): 7.6% (source: Vals AI boards, read October 21, 2026) - Code migration (Vals AI): 4.9% (source: Vals AI boards, read October 21, 2026) - ProgramBench (Vals AI): 0.0% (source: Vals AI boards, read October 21, 2026) - Terminal-Bench 2.1 (Vals AI): 50.9% (source: Vals AI boards, read October 21, 2026) - Finance Agent (Vals AI): 37.7% (source: Vals AI boards, read October 21, 2026) - Legal Research Bench (Vals AI): 15.4% (source: Vals AI boards, read October 21, 2026) - Excel Modeling Benchmark (Vals AI): 31.7% (source: Vals AI boards, read October 21, 2026) - LegalBench (Vals AI): 82.1% (source: Vals AI boards, read October 21, 2026) - CorpFin (Vals AI): 65.5% (source: Vals AI boards, read October 21, 2026) - TaxEval (Vals AI): 73.1% (source: Vals AI boards, read October 21, 2026) - MedCode (Vals AI): 38.6% (source: Vals AI boards, read October 21, 2026) - GPQA Diamond (Artificial Analysis run): 86.7% (source: Artificial Analysis, read October 21, 2026) - Humanity's Last Exam (Artificial Analysis run): 28.4% (source: Artificial Analysis, read October 21, 2026) - SciCode: 40.3% (source: Artificial Analysis, read October 21, 2026) - Terminal-Bench Hard (Artificial Analysis run): 36.4% (source: Artificial Analysis, read October 21, 2026) - IFBench: 81.4% (source: Artificial Analysis, read October 21, 2026) - Long context reasoning (AA-LCR): 79.3% (source: Artificial Analysis, read October 21, 2026) - τ²-Bench Telecom: 83.3% (source: Artificial Analysis, read October 21, 2026) - FrontierCode: 13.6% (source: Epoch AI benchmarking hub, read October 21, 2026) - WeirdML: 43.5% (source: Epoch AI benchmarking hub, read October 21, 2026) - APEX-Agents: 22.7% (source: Epoch AI benchmarking hub, read October 21, 2026) - GPQA Diamond (Epoch AI run): 85.4% (source: Epoch AI benchmarking hub, read October 21, 2026) - Mystery game puzzles: 20.0% (source: Epoch AI benchmarking hub, read October 21, 2026) - OTIS Mock AIME: 86.7% (source: Epoch AI benchmarking hub, read October 21, 2026) - ProofBench: 2.0% (source: Epoch AI benchmarking hub, read October 21, 2026) - DTBench: 90.1% (source: Epoch AI benchmarking hub, read October 21, 2026) - LMCA: 36.9% (source: Epoch AI benchmarking hub, read October 21, 2026) - Epoch Capabilities Index: 146.0 (source: Epoch AI benchmarking hub, read October 21, 2026) - LMArena Text: 1445 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: coding: 1468 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: hard prompts: 1452 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: math: 1457 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: instruction following: 1418 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: creative writing: 1399 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: English: 1455 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: non-English: 1427 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: Chinese: 1497 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: German: 1472 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: Korean: 1384 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: French: 1472 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: Spanish: 1455 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: Russian: 1417 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: Polish: 1385 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: multi-turn: 1411 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: longer queries: 1437 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: expert: 1472 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: hard prompts in English: 1453 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: software and IT: 1468 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: mathematical work: 1461 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: science: 1465 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: medicine and healthcare: 1456 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: legal and government: 1452 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: business and finance: 1434 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: writing and language: 1406 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026) - LMArena Text: entertainment and media: 1397 Elo (source: LMArena (Hugging Face dataset), read October 21, 2026)