LLMs TigerLive LLM benchmarks

nemotron-3-ultra: benchmark scores, price and ranking

nemotron-3-ultra from Nvidia ranks #473 of 743 LLMs on the LLMs Tiger aggregate benchmark score (47.0), based on 13 benchmark results captured up to 2026-10-13.

Organisation
Nvidia
Type
Open-weight / local model
Open weights
Yes
Released
2026-06-04 (source: epoch)
Cheapest API price
$0.600 in / $2.40 out per million tokens (via openrouter, observed 2026-10-13T19:47Z)
Aggregate rank
#473 of 743 (score 47.0)

Relative to other models, nemotron-3-ultra is strongest on Epoch ECI (#24 of 204) and weakest on ProofBench (#44 of 46).

Across the 13 benchmarks where at least ten models are scored, nemotron-3-ultra is in the top 10% on 0 and in the top half on 7.

Its cheapest listed input price, $0.600 per million tokens, is lower than that of 44% of the 427 models with a listed price.

nemotron-3-ultra benchmark results

nemotron-3-ultra benchmark scores, ranked against all models measured on each benchmark
BenchmarkScoreRankSourceCaptured
Epoch ECI146#24 of 204Epoch AI Benchmarking Hub2026-10-13
DTBench90.1%#29 of 146Epoch AI Benchmarking Hub2026-10-13
OTIS Mock AIME86.7%#52 of 189Epoch AI Benchmarking Hub2026-10-13
GPQA Diamond85.4%#78 of 275Epoch AI Benchmarking Hub2026-10-13
LMCA36.9%#40 of 108Epoch AI Benchmarking Hub2026-10-13
WeirdML43.5%#47 of 109Epoch AI Benchmarking Hub2026-10-13
CritPt3.1%#57 of 115Epoch AI Benchmarking Hub2026-10-13
Chess Puzzles12.0%#64 of 122Epoch AI Benchmarking Hub2026-10-13
Mystery Games20.0%#34 of 59Epoch AI Benchmarking Hub2026-10-13
SciCode40.3%#63 of 106Epoch AI Benchmarking Hub2026-10-13
FrontierCode13.6%#12 of 15Epoch AI Benchmarking Hub2026-10-13
APEX-Agents22.7%#15 of 16Epoch AI Benchmarking Hub2026-10-13
ProofBench2.0%#44 of 46Epoch AI Benchmarking Hub2026-10-13

About nemotron-3-ultra

nemotron-3-ultra is an open-weight model that can run on your own hardware from Nvidia. It was released on 2026-06-04, according to Epoch AI. The cheapest API listing we track charges $0.600 per million input tokens and $2.40 per million output tokens, through openrouter. We hold 13 benchmark results for it across 4 areas, and none is in the top tenth of its board.

Where nemotron-3-ultra stands

On our aggregate score, nemotron-3-ultra is #473 of 743 ranked models, in the lower half. Its score is 47.0; the scale is centred on 50, so it sits 3.0 points below the typical model. That beats 36% of the ranked field. Among the 2 ranked Nvidia models it is #1. The next one down is Nemotron 3 Super at 45.3.

nemotron-3-ultra, benchmark by benchmark

Epoch ECI puts nemotron-3-ultra at 146, #24 of 204. That beats 88% of the models tested, but 10 points short of Qwen 3.8 Max (156). The results either side are DeepSeek-V3.2 (146) and Kimi K2 Thinking (146). It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so nemotron-3-ultra is 19 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On DTBench, nemotron-3-ultra reaches 90.1%. It ranks #29 of 146, ahead of 80% of the field, and 8.8 points behind Claude Opus 5.5 (max) at 98.9%. The results either side are grok-4-20 (90.1%) and Claude Sonnet 4.6 (max) (89.9%). It covers dynamic tool-use benchmark. The board median is 68.8%, so nemotron-3-ultra is 21.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

nemotron-3-ultra manages 86.7% on OTIS Mock AIME, good for #52 of 189. Claude Fable 5.1 (max) leads at 100.0%, 13.3 points further up, and 72% of models sit below. The results either side are GPT-5 mini (high) (86.7%) and Qwen3-235B-A22B-Thinking-2507 (86.7%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so nemotron-3-ultra is 28.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPQA Diamond puts nemotron-3-ultra at 85.4%, #78 of 275. That beats 72% of the models tested, but 10.4 points short of GPT-6 Astra (max) (95.8%). The results either side are Qwen3.5 27B (85.5%) and GLM-5-FP8 (85.4%). It covers expert-level science questions. The board median is 71.7%, so nemotron-3-ultra is 13.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On LMCA, nemotron-3-ultra reaches 36.9%. It ranks #40 of 108, ahead of 63% of the field, and 31.3 points behind Claude Opus 5.5 (max) at 68.2%. The results either side are Claude Opus 4.1 (unknown thinking) (37.1%) and GPT-5.4 nano (xhigh) (36.9%). It covers language-model capability assessment. The board median is 29.2%, so nemotron-3-ultra is 7.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

nemotron-3-ultra manages 43.5% on WeirdML, good for #47 of 109. GPT-6 Astra (max) leads at 93.3%, 49.8 points further up, and 57% of models sit below. The results either side are o3-mini (high) (43.7%) and Mercury 2 (43.2%). It covers unusual machine-learning tasks. The board median is 41.2%, so nemotron-3-ultra is 2.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

nemotron-3-ultra lands below the middle on SciCode: 40.3%, #63 of 106, with 58% of the field scoring higher. Claude Opus 5.5 (max) leads at 66.9%; the closest result above is GPT-4.1 mini at 40.4%. It covers scientific research code. The board median is 44.3%, so nemotron-3-ultra is 4.1 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

nemotron-3-ultra lands below the middle on FrontierCode: 13.6%, #12 of 15, with 73% of the field scoring higher. Claude Opus 5 (max) leads at 53.4%; the closest result above is MiniMax-M3 at 14.7%. It covers hard programming tasks. The board median is 31.8%, so nemotron-3-ultra is 18.2 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

APEX-Agents is a weak spot. nemotron-3-ultra scores 22.7%, #15 of 16, behind 88% of the models measured. The leader, Claude Sonnet 5.5 (max), reaches 75.5%. It covers long professional agent tasks. The board median is 47.8%, so nemotron-3-ultra is 25.1 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

ProofBench is a weak spot. nemotron-3-ultra scores 2.0%, #44 of 46, behind 93% of the models measured. The leader, Claude Fable 5.1 (max), reaches 100.0%. It covers writing mathematical proofs. The board median is 21.0%, so nemotron-3-ultra is 19.0 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

The full table above also lists the other 3 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: uneven, ahead of 30% of the field on average across 2 benchmarks. Its best board here is SciCode (#63 of 106) and its weakest is FrontierCode (#12 of 15). Reasoning, maths and science: uneven, ahead of 50% of the field on average across 4 benchmarks. Its best board here is OTIS Mock AIME (#52 of 189) and its weakest is ProofBench (#44 of 46). Agents and tool use: uneven, ahead of 43% of the field on average across 2 benchmarks. Its best board here is DTBench (#29 of 146) and its weakest is APEX-Agents (#15 of 16). General knowledge and chat: solid, ahead of 60% of the field on average across 5 benchmarks. Its best board here is Epoch ECI (#24 of 204) and its weakest is Mystery Games (#34 of 59).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so nemotron-3-ultra costs more than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on Llama-4-Maverick-17B-128E-Instruct ($0.630) and Kimi K2.6 ($0.650).

Release history

The Nvidia model released before it was Nemotron 3 Super on 2026-03-11, which scores 45.3.

Reading these numbers

The 13 scores on this page come from 1 source: Epoch AI Benchmarking Hub (13). No official leaderboard supplied any of them, so treat them as third-party measurements.

Models ranked nearby

More from Nvidia

Compare it on the live leaderboard → · How the aggregate score works