Upstage / upstage/solar-pro4

Solar Pro 4 - access through LLMTR

Solar Pro 4 is Upstage's flagship model, tuned for agentic work. It is markedly stronger than Solar Pro 3 on document-based tasks, coding, and business workloads. Its 524,288-token context window holds a large codebase or a stack of long documents in one session, and it can emit up to 128K tokens in a single response — the only separate output ceiling in the family. Reasoning is off by default. You select a level with `reasoning_effort` across none, minimal, low, medium, high, xhigh, and max, but the model only begins thinking at `low`: `none` and `minimal` are accepted, return 200, and produce no reasoning tokens at all. When reasoning is on, the trace is returned separately in the response's `reasoning` field and is streamed incrementally as well. It supports native tool/function calling, and every `tool_choice` value works — auto, required, none, and a named choice. It produces JSON object output and structured output that conforms exactly to a supplied schema. Repeating a prompt prefix engages the prompt cache and those tokens are billed at the discounted rate. It accepts text only; image and audio input are not supported. Official language coverage is English, Korean, and Japanese. Note that reasoning tokens are drawn from the `max_tokens` budget, so keep it at 200-250 or above while reasoning is on — otherwise the response comes back with `finish_reason: length` and empty content.