NVIDIANemotron 3 Ultra 550B A55B
Best configuration ranks #124 of 610 on the BenchLeader Index at 58.3 ±6.9. Last measured 5 Sept 2026. Released 4 Jun 2026.
- Blended price
- $1.00/M
- $0.500 in · $2.50 out
- Output speed
- 156 tok/s
- First answer
- 17 s
- first token 0.40 s
- Context
- 1M
- Overall index58
- Coding62
- Agents & tools53
- Knowledge59
- Instruction following78
- Long context66
- Composite43
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | default | Source |
|---|---|---|
| LiveBench Reasoningnot in index | 74.7%#47 | LiveBench |
| GPQA Diamond (AA)not in index | 86.7%#107 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 28.4%#117 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | 86.1%#42 | Vals AI |
Coding
| Benchmark | default | Source |
|---|---|---|
| LiveBench Codingnot in index | 70.7%#47 | LiveBench |
| SciCode (AA)not in index | 40.3%#113 | Artificial Analysis |
| LiveCodeBench | 86.0%#23 | Vals AI |
| SWE-bench (Vals)not in index | 69.0%#64 | Vals AI |
Agents & tools
| Benchmark | default | Source |
|---|---|---|
| LiveBench Agentic Codingnot in index | 38.7%#51 | LiveBench |
| Terminal-Bench Hard | 36.4%#69 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 83.3%#106 | Artificial Analysis |
| Terminal-Bench 2.1 (Vals) | 50.9%#47 | Vals AI |
Maths
| Benchmark | default | Source |
|---|---|---|
| LiveBench Mathematicsnot in index | 88.7%#29 | LiveBench |
Knowledge
| Benchmark | default | Source |
|---|---|---|
| LiveBench Data Analysisnot in index | 54.5%#50 | LiveBench |
| AA-Omniscience | -0.4#97 | Artificial Analysis |
| MMLU-Pro | 85.8%#53 | Vals AI |
| LegalBench | 82.1%#68 | Vals AI |
| CorpFin | 65.5%#27 | Vals AI |
| TaxEval | 73.1%#52 | Vals AI |
Instruction following
| Benchmark | default | Source |
|---|---|---|
| LiveBench Languagenot in index | 70.8%#49 | LiveBench |
| IFBench | 81.4%#4 | Artificial Analysis |
Long context
| Benchmark | default | Source |
|---|---|---|
| AA-LCR | 79.3%#65 | Artificial Analysis |
Composite
| Benchmark | default | Source |
|---|---|---|
| LiveBench | 67.4%#47 | LiveBench |
| AA Intelligence Index | 23.4#132 | Artificial Analysis |
| Vals Indexnot in index | 27.4#43 | Vals AI |
Where it wins
Benchmarks where this configuration ranks in the top five of every configuration measured.
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Baseten (US) | 124 tok/s | 0.40 s | $0.600 | $2.40 | 203k | fp4 |
| Baseten | 121 tok/s | 0.38 s | $0.600 | $2.40 | 203k | fp4 |
| Venice | 63 tok/s | 4.47 s | $0.625 | $3.13 | 256k | fp8 |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.150 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0010 | $0.0008 | 18.8 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0075 | $0.0044 | 20.7 s |
| Code edit | 6,000 / 1,500 | $0.0068 | $0.0052 | 26.5 s |
| Agentic coding session | 60,000 / 4,000 | $0.040 | $0.024 | 42.5 s |
| Structured extraction | 2,000 / 200 | $0.0015 | $0.0010 | 18.2 s |
See also
Data as of 9 Sept 2026. Compare with another model.