Solar Mini 4 API: Upstage's small MoE with a 524K context

Upstage's Solar Mini 4 is on FastMetal as solar-mini4. It has the same 524,288-token context as Solar Pro 4 and costs less per token, which makes it a fit for high-volume work with short inputs and outputs.

What it is

• A small mixture-of-experts model: 35 billion parameters in total, 3 billion active per token. Upstage pitches it at agent workloads where latency and cost matter.
• A 524,288-token context, the same as Solar Pro 4.
• Tool calling and structured output. We confirmed both tool calls and json_schema output in real calls through FastMetal.
• Text input only. It does not accept images.
• It answers in Japanese as well as English. Asked "日本の首都は？一言で。" ("Capital of Japan? One word."), it replied "東京です。" ("Tokyo.").

Calling it

Send solar-mini4 as the model to the OpenAI-compatible endpoint. Classification is a typical job for it:

curl https://api.fastmetal.ai/v1/chat/completions \
  -H "Authorization: Bearer $FASTMETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "solar-mini4",
    "messages": [
      {"role": "user", "content": "Classify this support ticket as returns, shipping or other: The item I received is the wrong size."}
    ]
  }'

Classification and extraction, many short calls in a row, are where a low per-token rate adds up.

How it is billed

Solar Mini 4 is billed per million input and output tokens. Its rates follow Upstage's own pricing, and we revise them when Upstage does. Current rates are on the Solar Mini 4 model page and the pricing page.

Billing is prepaid: you load a balance by card, from $5, and requests stop when it runs out.

Solar Mini 4 or Solar Pro 4

The comparison page puts the two models' answers to the same prompts side by side, so you can judge whether the larger model earns its higher rate for your task.

Everything else we serve is in the model catalog.