PricePerToken lists Nvidia’s Llama 3.3 Nemotron Super 49B V1.5 at $0.40 per million input tokens and $0.40 per million output tokens, while Nvidia’s own model card describes the system as a Llama 3.3 70B Instruct derivative with a 131,072-token context window and preview API access.
Nvidia’s Llama 3.3 Nemotron Super 49B V1.5 has been listed by PricePerToken with API pricing of $0.40 per million input tokens and $0.40 per million output tokens, according to a model catalog page updated June 21, 2026.
The listing identifies the model as part of Nvidia’s Nemotron family and highlights a 131K-token context window, alongside benchmark results for MMLU Pro, GPQA, LiveCodeBench, AIME, math, and general intelligence evaluations. PricePerToken’s entry is a third-party catalog, so its pricing should be read as a cataloged API price rather than a new Nvidia announcement.
Nvidia’s own model card for Llama-3.3-Nemotron-Super-49B-v1.5 describes the model as a derivative of Llama-3.3-70B-Instruct. The company lists the release date as July 25, 2025, the version as 1.5, and the context length as 131,072 tokens.
The Nvidia page also indicates preview API access. That matters for developers evaluating the model, because preview availability can differ from a fully generally available production service in terms of access terms, stability expectations, or deployment options.
A separate Nvidia NIM API reference describes the model as a reasoning large language model intended for retrieval-augmented generation, tool calling, and chat. Nvidia’s API documentation lists Nvidia as the developer and says the model was trained between November 2024 and July 2025. The API reference describes the context length as 128K, which is consistent in practical terms with the 131,072-token figure stated in Nvidia’s model card.
The combination of long context and per-token pricing is a practical consideration for teams building applications around retrieval, agentic workflows, support automation, code assistance, or long-document analysis. A 131,072-token context window allows substantially more source material to be sent in a single request than standard short-context models, though actual cost depends on how many input and output tokens an application uses.
At the cataloged $0.40 per million input tokens and $0.40 per million output tokens, PricePerToken positions the model in a pricing format that is directly comparable with other API-hosted language models. However, the catalog excerpt does not specify whether pricing varies by provider, region, deployment method, throughput tier, or preview status. Buyers should confirm final commercial terms with the serving platform before relying on the listing for procurement decisions.
PricePerToken says its page includes detailed benchmark scores for MMLU Pro, GPQA, LiveCodeBench, AIME, math, and intelligence. Those benchmarks cover different capabilities: knowledge and reasoning, graduate-level question answering, coding, competition-style math, and broader evaluation categories.
The provided source excerpts do not include the actual numeric scores, benchmark methodology, evaluation settings, or comparisons against other models. For that reason, the safest reading is that the catalog reports benchmark coverage, not that the model has achieved a particular ranking or superiority over competing systems.
Nvidia’s NIM documentation frames Llama 3.3 Nemotron Super 49B V1.5 as a reasoning model for RAG, tool use, and chat. That positioning aligns with current enterprise use cases in which a model must combine instruction following, external information retrieval, and structured interactions with tools or APIs.
The model’s base lineage is also notable. Nvidia’s model card says it is derived from Llama-3.3-70B-Instruct, while the name indicates a 49B-parameter Nemotron “Super” variant. The provided sources do not explain the compression, distillation, pruning, or optimization methods used to arrive at the 49B variant, so those details should not be inferred from the listing alone.
For developers, the key sourced facts are clear: Nvidia identifies the model as version 1.5 with a July 2025 release date and 131,072-token context; Nvidia’s NIM API reference describes it as a reasoning LLM for RAG, tool calling, and chat; and PricePerToken’s June 2026 catalog page lists API pricing at $0.40 per million input and output tokens.
PricePerToken’s entry is a third party catalog, so its pricing should be read as a cataloged API price rather than a new Nvidia announcement.
What Nvidia’s model card says Nvidia’s own model card for Llama 3.3 Nemotron Super 49B v1.5 describes the model as a derivative of Llama 3.3 70B Instruct.
The company lists the release date as July 25, 2025, the version as 1.5, and the context length as 131,072 tokens.
Continue reading
