Solar Mini 4

Upstage's compact 35B-parameter mixture-of-experts model with 3B active parameters and a 524K context window. Built for fast, cost-efficient agentic tasks, with strong Korean and English support.

Tool Calling
Structured Output

Added Sep 23, 2026

Pricing

Auto routing · per 1M tokens

Input: $0.050
Output: $0.20
Cache read: $0.0050

Compare provider prices

Specifications

Context window: 524.3K
Max output: 131.1K
Parameters: 35B / 3B; Total / active
Avg output (7d): 61 tokens; Longer than 5% of models

Benchmarks

Artificial Analysis

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

24.1

Better than 76% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

22.3%

Better than 43% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

871 Elo

Better than 38% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

1090 Elo

Better than 52% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

5.0%

Better than 24% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

83.3%

Better than 96% of models compared

Reasoning

HLE

Humanity's Last Exam

25.8%

Better than 73% of models compared

CritPt

Research-level physics reasoning

1.4%

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

1.0%

Better than 42% of models compared

SciCode

Python programming for scientific computing

47.6%

Better than 45% of models compared

Knowledge

AA-Omniscience Accuracy

Proportion of correctly answered questions

18.4%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

35.8%

Legacy benchmarks

AA-LCR (unversioned / legacy)

Long context reasoning evaluation

83.3%

Better than 96% of models compared

GDPval-AA (unversioned / legacy)

Economically valuable tasks

29.5%

Last updated Oct 14, 2026