Nemotron 3 Ultra
- Input
- $0.60/M tok
- Output
- $2.50/M tok
- Blended
- $1.075/M tok · 3:1
- Speed
- 155tok/s
- Context
- 262Ktokens
- Cost per task
- ~$1.1/taskest.
Open weights (550B-A55B). Price is the median third-party host price.
Rankings
Coding
Mixed#78 / 137rank by quality40.5±5.4quality / 1004/7 benchmarks measuredWriting & chat
Verified#59 / 162rank by quality58.6±3.3quality / 1004/5 benchmarks measuredHard reasoning
Verified#75 / 218rank by quality55.4±3.7quality / 1003/4 benchmarks measuredAgents
Mixed#98 / 187rank by quality30.2±3.4quality / 1005/7 benchmarks measured
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.
- Hardware
- 8× B200 180GB
- FP8 weights
- Throughput
- ~21,434 tok/s
- many requests batched
- Price
- $0.29–2.61 /M tok
- busy → light use
- On your own machine
- Mac Studio M5 Ultra 512GB (int4, ~31 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (550B params, 55B active per token) + 25% KV-cache headroom ≈ 688 GB
- 8× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~21,434 output tok/s aggregate at batch 613 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~154,636 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Coding
SWE-bench VerifiedLab-reported
#16 of 2070.7%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #15Nemotron 3.5 Super VL NVFP471.1%
- #16Nemotron 3 Ultra70.7%
- #17Intern-S2-Preview (35B-A3B)64.0%
huggingface.co · 2026-06-04
SWE-bench Pro (Public)Estimated
~#23 of 24 (estimated)~50.4%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #22Muse Glimmer51.2%
- ~#23Nemotron 3 UltraEstimated50.4%
- #23Nex-N2.5-Mini43.8%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#40 of 46 (estimated)~42.7%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #39Beam 1.0 501B A23B43.3%
- ~#40Nemotron 3 UltraEstimated42.7%
- #40Qwen3.8 27B42.2%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Verified
#53 of 1010.5%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #53Muse Glimmer0.5%
- #53Nemotron 3 Ultra0.5%
- #53Nemotron 3.5 Lightning0.5%
artificialanalysis.ai · 2026-09-24
LiveCodeBenchLab-reported
#5 of 2789.0%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #4Ternary Bonsai 2 27B90.1%
- #5Nemotron 3 Ultra89.0%
- #6Beam 1.0 501B A23B88.3%
huggingface.co · 2026-06-04
SciCodeVerified
#78 of 11840.3%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #77Trinity Large Thinking40.6%
- #78Nemotron 3 Ultra40.3%
- #79Mistral Medium 3.540.2%
artificialanalysis.ai · 2026-09-24
AA Coding Agent IndexEstimated
~#25 of 26 (estimated)~30.8
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #24GPT-6 Luna31.1
- ~#25Nemotron 3 UltraEstimated30.8
- #25Muse Spark Open26.3
No published score — estimated from related benchmarks
Writing & chat
IFBenchVerified
#4 of 13981.4%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #3Qwen3.8-Omni-Flash81.5%
- #4Nemotron 3 Ultra81.4%
- #5Qwen3.8 Flash-Next81.3%
artificialanalysis.ai · 2026-09-24
LMArena Text, Non-English (Elo)Verified
#38 of 561405 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #37Mistral Medium 3.51406
- #38Nemotron 3 Ultra1405
- #39Mistral Large 31397
lmarena.ai · 2026-09-24
LMArena Text (Elo)Verified
#36 of 561426 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #36Mistral Medium 3.51426
- #36Nemotron 3 Ultra1426
- #38Muse GlimmerFew votes1424
lmarena.ai · 2026-09-24
#16 of 251692 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #15Gemini 3.8 Flash1748
- #16Nemotron 3 Ultra1692
- #17Qwen3.8 27B1671
eqbench.com · 2026-09-24
Hard reasoning
Humanity's Last ExamVerified
#61 of 19828.4%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #60Solar Open2 250B28.5%
- #61Nemotron 3 Ultra28.4%
- #62GPT-5.4 nano28.3%
artificialanalysis.ai · 2026-09-24
AIME (latest)Estimated
~#12 of 34 (estimated)~89.9%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #11Xing4.0-29B-A4B90.0%
- ~#12Nemotron 3 UltraEstimated89.9%
- #12gpt-oss-20b89.3%
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Verified
#26 of 11983.3%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #26GPT-5.4 mini83.3%
- #26Nemotron 3 Ultra83.3%
- #28Nex-N2-Pro81.6%
artificialanalysis.ai · 2026-09-24
OSWorld-VerifiedEstimated
~#12 of 15 (estimated)~60.2%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #11Muse Glimmer65.9%
- ~#12Nemotron 3 UltraEstimated60.2%
- #12Nemotron 3.5 Super VL NVFP452.3%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
No published score — estimated from related benchmarks
BrowseCompLab-reported
#20 of 2144.4%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #19Mistral Medium 3.548.6%
- #20Nemotron 3 Ultra44.4%
- #21Mistral Small 421.3%
huggingface.co · 2026-06-04
GDPval-AA (Elo)Verified
#41 of 911013 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #40Kimi K2.7 Code1025
- #41Nemotron 3 Ultra1013
- #42K2 Horizon 7B1011
artificialanalysis.ai · 2026-10-10