Upstage: Solar Pro 3

Solar Pro 3 is a text generation model from Upstage, released on 27 January 2026. It is a Mixture-of-Experts (MoE) language model with 102B total parameters and 12B active parameters per forward pass, which means only a fraction of the network is engaged for each token it generates. That design is intended to keep inference computationally efficient while still drawing on a large overall parameter pool. It sits in the text category, so it is aimed at language tasks rather than image, audio or video work.

The model accepts text as input and returns text as output. It has a context window of 131,072 tokens, and its maximum output length is 117,964 tokens, so it can produce very long responses in a single call. Supported request parameters include temperature, top_p, frequency_penalty, presence_penalty and max_tokens for controlling sampling and length. It also supports response_format and structured_outputs, which let you constrain the shape of the returned text. On the reasoning side, the model exposes reasoning, reasoning_effort and include_reasoning, and it supports tool use through tools, tool_choice and parallel_tool_calls.

Those capabilities point to a few practical patterns. The large context window suits long document work: summarising lengthy reports, answering questions over a full codebase or contract, or keeping a long conversation coherent without trimming history. Structured outputs and response_format make it a reasonable fit for extraction pipelines where you need JSON or another fixed schema rather than free prose. Tool calling and parallel tool calls support agent-style workflows, where the model decides which functions to invoke and can issue several calls at once. The reasoning controls are useful when you want the model to work through a problem step by step, or when you want to dial that effort up or down depending on the task. The large output ceiling helps for tasks that generate long artefacts, such as drafting documentation or expanding an outline into full sections.

You call it through api14's OpenAI-compatible API using the model id "upstage/solar-pro-3". Because the interface follows the OpenAI chat completions shape, you can point an existing OpenAI client at api14's base URL and set the model field to that id, without rewriting your integration. The request body accepts the parameters listed above, including temperature, top_p, max_tokens, frequency_penalty, presence_penalty, response_format, structured_outputs, tools, tool_choice, parallel_tool_calls, reasoning, reasoning_effort and include_reasoning. Live pricing for the model is shown in the price table on this page.

One thing to keep in mind is that Solar Pro 3 is text only. If your task depends on image, audio or video input, this model will not handle it, and you would need a multimodal model instead. The same applies if you need a non-text output modality. For text workloads within the 131,072 token context window, though, the combination of long context, structured output support and tool calling covers a wide range of applications.