Nous: Hermes 3 70B Instruct

Nous: Hermes 3 70B Instruct is a text generation model from Nous Research, built on the Llama 3.1 70B base. It is the third generation of the Hermes line, and Nous describes it as a generalist language model with improvements over Hermes 2, including advanced agentic capabilities, better roleplaying, reasoning, multi-turn conversation, and long context coherence. The model was released by the provider on 18 August 2024, and its knowledge cutoff is 31 December 2023.

The model accepts text input and returns text output, so it is a pure language model with no image or audio handling. Its context window is 131,072 tokens, which is large enough to hold long documents, extended conversations, or sizeable code files in a single request. The maximum output is 16,384 tokens, so responses can be long, but they are capped well below the input context size. Because the context is long and the model is tuned for coherence across it, it can be used for tasks where earlier parts of a conversation or document need to stay relevant many turns later.

Those capabilities point to a few practical uses. Agentic workflows are a stated strength, so tool calling and multi step task orchestration are reasonable fits. The roleplaying and multi-turn conversation improvements suit chat assistants, character driven applications, and interactive experiences that need consistent persona and memory over a long session. Reasoning and long context coherence make it usable for document question answering, summarisation of lengthy material, and code assistance where a whole file or repository section must be considered at once. Structured outputs are also supported, which helps when the model must return data in a fixed shape for downstream parsing.

You call the model through api14's OpenAI-compatible API using the model id "nousresearch/hermes-3-llama-3.1-70b". Because the endpoint follows the OpenAI chat completions format, existing OpenAI SDKs and clients work with only a base URL and key change. The request parameters supported are frequency_penalty, logit_bias, max_tokens, min_p, presence_penalty, repetition_penalty, seed, stop, structured_outputs, temperature, top_k, and top_p. The seed parameter allows reproducible sampling, and structured_outputs constrains generation to a defined schema. Note that min_p, repetition_penalty, and top_k are not part of the standard OpenAI parameter set, so check that your client passes them through rather than dropping unknown fields.

The main constraints to keep in mind are the text only input and output, the 16,384 token output ceiling, and the knowledge cutoff of 31 December 2023, so anything after that date is outside the model's training data. If your task depends on image or audio input, this model will not fit, and you would need a multimodal model instead. For work that fits within text and the stated context and output limits, Hermes 3 70B Instruct is a straightforward choice through the api14 gateway.