Upstage
Upstage

Solar Mini 4

Release Date 2026-09-23(Knowledge cutoff 2026-02)524K IN · 131K OUTProprietary ModelAPI available
OverallNo data
AutomationNo data
Docs & office4.0Top 16%Coding3.0Top 42%

Solar Mini is Upstage's compact language model for speed- and cost-sensitive work, supporting Korean, English and Japanese documents and tool workflows. It shares long-context support, reasoning, structured outputs and tool calling with the same-generation Pro, but matching feature lists do not imply equal accuracy on difficult tasks. AIB places document work in the upper band at the time checked, supporting evaluation of extraction and organization through omission rates and correction time on real files. Upstage offers API and separate on-premises routes; verify deployment contracts rather than treating on-premises availability as freely published weights. For high-volume use, compare reasoning latency, output volume and cache behavior, and validate document quality instead of assuming a large context guarantees complete understanding.

API cost

Provider price
$0.05IN$0.2OUT
SourceOpenRouterLast updated

Real-world usage

No.303 places0.4% share · 1.0T tokens/wk
384B9/28
932B10/5
1.0T10/12
1.2T10/19
1.4T10/26
1.4T11/2
1.3T11/9
1.0T11/16
SourceOpenRouterLast updated

Output Speed

Fastper 2,000 tokens
91tok/s
First Output 1.7stotal: 23.7s
SourceOpenRouterLast updated

User Comparisons

Writing Style
Answer length
Shorter side
bottom 37%
Vocabulary diversity
Very plain
bottom 19%
Sentence rhythm
Steadier side
bottom 24%
Language mixing
High
top 40%
List usage
Often
18% of lines
Heading usage
Occasionally
11% of lines
Reasoning share
Minimal
0% of total
Response Speed
instability ±263%
First response 1.5sTotal 11.1s
63 tok/s
15.3ms per token
SourceAIB Inc.Last updated

Expert Performance Evaluation

Can it pull scattered facts from several long documents into one answer?Long Context ReasoningTop 12%#14
83.3%

More about Long Context Reasoning

1.0%47.6%

More about Terminal-Bench 4.0

Can it solve problems that stump experts across fields?Humanity's Last ExamBottom 35%#81
25.8%

More about Humanity's Last Exam

See each evaluation’s detail page for its source and last updateData corrections & removal requests