Solar Mini 4
Solar Mini 4 is Upstage's cost-efficient compact LLM for agentic use, with 35B/3B active parameters, 512K context, reasoning mode, and fluent Korean.
Model Overview
Capabilities, design details, and architectural traits
Solar Mini 4 - Upstage's compact model for high-volume agent workloads
Solar Mini 4 is a compact language model from Upstage, built for agentic use cases where response speed matters. It pairs trilingual fluency centered on Korean with a very long context window.
| Differentiator | Detail |
|---|---|
| 512K context, 128K output | Supports long-context workloads with up to 512K input tokens and 128K max output tokens. |
| Korean-first trilingual support | Fluent Korean alongside strong English and Japanese, for both input and output. |
| Full agentic capability set | Tool calling, structured outputs, and a dedicated reasoning mode are all supported. |
| On-premises deployment | Available on Upstage Console and Playground, and deployable on-premises for enterprise environments. |
| API privacy by default | Upstage does not collect API input data, store it, or use it for model training. |
Built for agents that run all day
The model's design targets agentic production: agents that call tools repeatedly, produce structured outputs, and need fast responses at scale. The reasoning mode adds multi-step problem solving on top of this workflow, so a single compact model can cover both quick tool-driven turns and harder reasoning steps.
Korean strength in a small package
Upstage positions Korean fluency as a core strength of the Solar family, and Solar Mini 4 carries it into the compact tier. This makes it a practical choice for Korean-language agent deployments.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Upstage | Anthropic | Anthropic |
| Release Date | September 22, 2026 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Feb 2026 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 524K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.10 Best Input Pricing | $4 | $2 |
| Output Pricing | $0.40 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 24.1 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Upstage
Other models by Upstage
Top AI Models
Leading alternatives by intelligence score
