Compare AI models
| Specifications | ||
ParametersThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. | 4B | 120B |
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 128k | 1M |
| API pricingUSD per 1M tokens · lower wins | ||
Cheapest inputLowest input rate across every provider on OpenRouter, the lab included. The cheapest endpoint may run a quantised build or a shorter context — see "Available from" on the model page. | — | $0.06DeepInfra |
Cheapest outputLowest output rate across every provider on OpenRouter, the lab included. May come from a different provider than the cheapest input. | — | $0.32DeepInfra |
| — | $35.00DeepInfra · bf16 | |
| Benchmarks | ||
AIME 2025Reasoning — single-answer accuracy on the 2025 invitational mathematics exam without tools | 46.3% | 90.21% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 55.1% | 79.23% |
BullshitBench v2Nonsense detection — Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better. | — | 54% |
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. | — | 60.47% |
SWE-Bench MultilingualMultilingual coding — Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better. | — | 45.78% |
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. | — | 81.19% |
Terminal-Bench 2.1Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better. | — | 39.61% |
Terminal-Bench 2.0Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? (Version 2.0 of the test.) Higher is better. | — | 31% |
IFEvalAgentic tasks — following verifiable instructions in generated responses | 75.5% | — |
τ²-bench · retailAgentic tasks — task success in the retail domain of the second-generation customer-service evaluation | — | 62.83% |
τ²-bench · airlineAgentic tasks — task success in the airline domain of the second-generation customer-service evaluation | — | 56.25% |
τ²-bench · telecomAgentic tasks — task success in the telecom domain of the second-generation customer-service evaluation | — | 64.36% |
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better. | — | 31.28% |
Humanity's Last Exam · text only, no toolsMultidisciplinary reasoning — accuracy on the 2,158 text-only questions without tools | — | 20.3% |
Humanity's Last Exam · text only, with toolsMultidisciplinary reasoning — The text-only questions from Humanity's Last Exam, answered with access to tools such as search or code execution. The score is the percentage answered correctly. | — | 22.82% |
MATH-500Reasoning — accuracy on the 500-problem MATH subset | 96.2% | — |
MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark | — | 83.73% |
AA Intelligence IndexOverall intelligence — Artificial Analysis composite intelligence index across evals | — | 36 |
| Overview | ||
| Company | NVIDIA | NVIDIA |
| Release date | May 20 2025 | Mar 11 2026 |
| Access | Open Weight | Open Weight |
| Model details | View model | View model |
Other comparisons
Llama Nemotron Nano 4B v1.1vsClaude Fable 5.2Nemotron 3 SupervsClaude Fable 5.2Llama Nemotron Nano 4B v1.1vsGPT-6.2 AstraNemotron 3 SupervsGPT-6.2 AstraLlama Nemotron Nano 4B v1.1vsGemini 4 NeonNemotron 3 SupervsGemini 4 NeonLlama Nemotron Nano 4B v1.1vsMuse EmberNemotron 3 SupervsMuse EmberLlama Nemotron Nano 4B v1.1vsGrok 4.7Nemotron 3 SupervsGrok 4.7Llama Nemotron Nano 4B v1.1vsDeepSeek-V4.1-Flash-1017Nemotron 3 SupervsDeepSeek-V4.1-Flash-1017Frequently asked questions
Nemotron 3 Super leads Llama Nemotron Nano 4B v1.1 on 2 of the 2 benchmarks they both report (AIME 2025, GPQA Diamond). Llama Nemotron Nano 4B v1.1 shipped 295 days before Nemotron 3 Super, so benchmark comparisons should account for the intervening progress.
Llama Nemotron Nano 4B v1.1 has 4B parameters, while Nemotron 3 Super has 120B. Context windows are 128k (Llama Nemotron Nano 4B v1.1) vs 1M (Nemotron 3 Super).
On AIME 2025, Nemotron 3 Super leads at 90.21% vs Llama Nemotron Nano 4B v1.1 at 46.3%. On GPQA Diamond, Nemotron 3 Super leads at 79.23% vs Llama Nemotron Nano 4B v1.1 at 55.1%.