Overview
Solar Mini 4 is a compact language model from Upstage, released September 22, 2026. Upstage describes it as its cost-efficient, compact model for agentic use cases where response speed and cost matter. It has 35B total parameters with 3B active, and pairs fluent Korean with strong English and Japanese.
The model takes a 512K-token context and can write up to 128K output tokens. Upstage lists chat, reasoning, structured outputs and tool calling as its capabilities, with a training-data cut-off of February 2026. It is available in the Upstage Console, in the Playground and as an on-premises deployment; on the API it is called through the stable alias solar-mini4, which points to the snapshot solar-mini4-260922.
Upstage released Solar Jev on the same day as a structured decision model served on Solar Mini 4. Through the POST /v1/systemone endpoint it takes a state plus typed questions and returns a choice, a score or a yes/no answer, each with a probability taken directly from the model rather than written out as text, which Upstage positions for routing, classification and policy checks.
Upstage did not publish a benchmark table for Solar Mini 4 at release, so this page lists its specifications and pricing only.
| Released | 2026-09-22 |
|---|---|
| License | Proprietary (Upstage API and on-premises) |
| Weights | API only |
| Parameters | 35B total · 3B active |
| Context | 512K tokens |
| Max output | 128K tokens |
| Knowledge cutoff | February 2026 |
| Modalities | Text |
| Status | Generally available |
Pricing
| Input | $0.10 / 1M tokens |
|---|---|
| Cached input | $0.01 / 1M tokens |
| Output | $0.40 / 1M tokens |
List price on the Upstage API. A launch promotion took 50% off ($0.05 in, $0.005 cached, $0.20 out) through October 22, 2026 (UTC). Prices exclude 10% VAT.
Strengths
- 35B total parameters with only 3B active, aimed at high-frequency, cost-sensitive agent workloads
- 512K-token context window with up to 128K output tokens
- Fluent Korean alongside English and Japanese
- Chat, reasoning, structured outputs and tool calling in one model
- Priced at $0.10 per 1M input tokens, $0.01 cached and $0.40 output
- Deployable on-premises as well as through the Upstage Console
Best for
- Reach for it when an agent makes many short, frequent calls and per-token cost and response speed matter more than peak capability.
- Reach for it when the workload is in Korean, or mixes Korean with English and Japanese.
- Reach for it when you need tool calling or structured outputs over long inputs of up to 512K tokens.
- Look elsewhere if you need image or audio input: Upstage's spec table lists chat, reasoning, structured outputs and tool calling, with no vision or audio capability.
How to access
| Provider | Model ID |
|---|---|
| Upstage API ↗ | solar-mini4 |
| OpenRouter ↗ | upstage/solar-mini4 |
FAQ
How big is Solar Mini 4?
Upstage lists Solar Mini 4 at 35B total parameters with 3B active. It has a 512K-token context window, can produce up to 128K output tokens, and has a training-data cut-off of February 2026.
How much does Solar Mini 4 cost?
On the Upstage API it costs $0.10 per million input tokens, $0.01 per million cached input tokens and $0.40 per million output tokens, excluding VAT. At launch Upstage ran a 50% discount through October 22, 2026 (UTC).
Which languages does Solar Mini 4 support?
Upstage lists English, Korean and Japanese, and describes the model as combining fluent Korean with strong English and Japanese.
What is Solar Jev?
Solar Jev is a structured decision model Upstage released on September 22, 2026, served on Solar Mini 4 through the POST /v1/systemone endpoint. It returns a choice, a score or a yes/no answer with a probability taken from the model, instead of generating prose.