NVIDIA: Llama 3.1 Nemotron Ultra 253B v1
Analysis Summary
Llama 3.1 Nemotron Ultra 253B v1 is NVIDIA's large reasoning-focused model built on the Llama 3.1 architecture, with a 131K context window and strong performance on math (index 63.7) and coding (livecodebench 0.641, AIME 0.637). The MMLU Pro score of 0.825 confirms broad academic knowledge depth.
However, the agentic index of just 6.8 and terminalbench hard score of 0.023 are very low, indicating poor reliability for multi-step tool use or autonomous workflows. The long-context retrieval score of 0.073 is also weak, limiting its usefulness for large document analysis despite the 131K window. Instruction following (ifbench 0.382) is below average.
At $0.60 input and $1.80 output, it offers reasonable value for pure math and coding tasks. Teams needing a strong mathematical reasoner or code generator for batch, non-agentic workloads will find it capable, but it should not be used in agentic or tool-orchestrated pipelines.
Assessed July 10, 2026
Editorial notes
NVIDIA Nemotron Ultra 253B has strong math and coding benchmarks including a livecodebench of 0.641, but very low agentic and long-context scores limit its business versatility.
Rankings consider pricing, capabilities, benchmarks, and real-world applicability and are refreshed as new models launch. Feedback?
DFO Verdict
NVIDIA Nemotron Ultra 253B has strong math and coding benchmarks including a livecodebench of 0.641, but very low agentic and long-context scores limit its business versatility.
Benchmark scores
Magenta = intelligence Ā· Ink = technical/agentic Ā· Cyan = content & long-context Ā· Grey = community benchmarks. Data: Artificial Analysis, Hugging Face.
9.1 Intelligence IndexĀ·6.8 Agentic IndexĀ·63.7 Math Index
How NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 compares
NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 ranks #251 of 393 AI models we track for overall intelligence, #273 of 300 for agentic tasks. Its 131K-token context window is larger than 57% of the models we list. At $0.60 per million input tokens it is cheaper than 34% of comparable models.
Dark bar = input Ā· light bar = output, scaled to the priciest peer.
1M tokens ā 8 full-length novels or ~2,500 pages of business documents in a single request.
Strongest on value. The pulled-in technical corner is the trade-off, and if the shape matters more than the price, this is your model.
Compare shapes side-by-side āPricing
| Token Type | Cost per 1M tokens | Cost per 1K tokens |
|---|---|---|
| Input | $0.60 | $0.000600 |
| Output | $1.80 | $0.001800 |
What would NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 cost your business?
Pick the job that looks most like yours, then fine-tune with the sliders. Estimates update live.
A website chatbot handling around 100 customer conversations a day, a few short messages each.
Full calculator with 612 models ā Price Calculator
These numbers get smaller with the right architecture.
We route routine calls to cheap models and save NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 for the hard ones. Most clients cut their estimate by 60-80%.
Talk to our teamAbout NVIDIA: Llama 3.1 Nemotron Ultra 253B v1
Llama-3.1-Nemotron-Ultra-253B-v1 is a large language model (LLM) optimized for advanced reasoning, human-interactive chat, retrieval-augmented generation (RAG), and tool-calling tasks. Derived from Metaās Llama-3.1-405B-Instruct, it has been significantly customized using Neural..
Embed this ranking
Frequently asked questions about NVIDIA: Llama 3.1 Nemotron Ultra 253B v1
How much does NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 cost?
NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 costs $0.60 per million input tokens and $1.80 per million output tokens.
What is the context window of NVIDIA: Llama 3.1 Nemotron Ultra 253B v1?
NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 has a context window of 131,072 tokens (131K).
Who created NVIDIA: Llama 3.1 Nemotron Ultra 253B v1?
NVIDIA: Llama 3.1 Nemotron Ultra 253B v1 is developed by NVIDIA and was released on April 8, 2025.
