Sao10K: Llama 3.1 Euryale 70B v2.2

Sao10K: Llama 3.1 Euryale 70B v2.2 is a text model from Sao10K, a provider known for fine-tunes aimed at creative writing and roleplay. It is built on Llama 3.1 70B and is the successor to Euryale L3 70B v2.1. The provider describes it as focused on creative roleplay, so it sits in the space of open-weight style models tuned for narrative and character driven conversation rather than general purpose assistant work.

The model takes text in and returns text out. It has a context window of 131072 tokens, which is large enough to hold long conversations, extended scenes or sizeable documents in a single request. The maximum output is 16384 tokens, so a single response can be quite long, though it is smaller than the input window. Its knowledge cutoff is 2023-12-31, which matters if your task depends on events after that date. Supported request parameters include temperature, top_p, top_k, min_p, repetition_penalty, frequency_penalty, presence_penalty, logit_bias, seed, stop, max_tokens and structured_outputs.

The practical uses follow from that profile. Creative roleplay and character chat are the clearest fit, since the model was tuned for them. Long form fiction, scene continuation and interactive storytelling also suit the large context window, because you can keep a lot of prior narrative in the prompt without truncating it. The sampling parameters give you fine control over tone and repetition, which helps when you want consistent character voice across a long session. Structured outputs can be useful when you want the model to return something more predictable than free prose, for example a scene summary or a set of fields inside a roleplay pipeline.

You call it through api14's OpenAI-compatible API using the model id "sao10k/l3.1-euryale-70b". Because the interface is OpenAI-compatible, you can point an existing OpenAI client at api14's base URL and pass that id in the model field. The request supports the parameters listed above, so you can set temperature, top_p, top_k, min_p, repetition_penalty, frequency_penalty, presence_penalty, logit_bias, seed, stop, max_tokens and structured_outputs as needed. A live price table on the page covers cost.

One limit to keep in mind is the knowledge cutoff of 2023-12-31, so anything requiring newer information will not come from the model itself. The output cap of 16384 tokens also means very long continuations need to be built across multiple calls. If your task is factual question answering, code generation or tool use rather than creative writing, a general purpose model may fit better.