Solar Mini 4

Upstage

Solar Mini 4 is Upstage's cost-efficient proprietary mixture-of-experts language model with 35B total and 3B active parameters, built for agentic workloads where speed and cost matter. It is text-only with English, Korean, and Japanese support, plus reasoning mode, tool calling, and structured outputs. Upstage Console docs list a 512K context window, up to 128K max output, and a February 2026 training cutoff, with list pricing of $0.10 input / $0.40 output / $0.01 cached per million tokens. The six catalog scores are third-party independent evaluations from Artificial Analysis (AA-LCR 83.3, SciCode 47.6, HLE text-only 25.8, Terminal-Bench 4.0 1.0, GDPval-AA v2.1 1072 Elo, AA-Briefcase v1.1 872 Elo).

Key Specifications

Parameters
35.0B
Context
524.3K
Release Date
September 22, 2026
Average Score
37.1%

Timeline

Key dates in the model's history
Announcement
September 22, 2026
Last Update
October 8, 2026
Today
October 22, 2026

Technical Specifications

Parameters
35.0B
Training Tokens
-
Knowledge Cutoff
February 1, 2026
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.10
Output (per 1M tokens)
$0.40
Max Input Tokens
524.3K
Max Output Tokens
131.1K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Other Tests

Specialized benchmarks
AA-Briefcase v1.1
Artificial Analysis; AA-Briefcase v1.1 Elo (max_score 3000). • Verified
29.1%
AA-LCR
Artificial Analysis independent evaluation; AA-LCR v1.1. • Verified
83.3%
GDPval-AA 2.1
Artificial Analysis; GDPval-AA v2.1 Elo (max_score 3000). • Verified
35.7%
Humanity's Last Exam (no tools, text-only)
Artificial Analysis; Humanity's Last Exam text-only. • Verified
25.8%
SciCode
Artificial Analysis independent evaluation; SciCode. • Verified
47.6%
Terminal-Bench 4.0
Artificial Analysis independent evaluation; Terminal-Bench 4.0. • Verified
1.0%

License & Metadata

License
proprietary
Announcement Date
September 22, 2026
Last Updated
October 8, 2026

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.