all the models — AI benchmark observatory
← Models

Solar Mini 4

Upstage · proprietary · solar-mini4

Solar Mini 4 is Upstage's cost-efficient proprietary MoE language model (35B total / 3B active) for agentic use where speed and cost matter. API id solar-mini4 (version solar-mini4-260922). Text→text with English, Korean, and Japanese; reasoning mode, tool calling, and structured outputs. Upstage Console docs: 512K context, up to 128K max output, February 2026 training cutoff. List pricing $0.10 input / $0.40 output / $0.01 cached per 1M tokens. Independent Artificial Analysis benches (not vendor self-report): 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. No separate Upstage-published bench table found beyond console/product claims; AA rows stored as third-party.

Not enough scores for a fingerprint yet.

Benchmark scores

BenchmarkScore
GDPval-AA 2.11072.00
AA-Briefcase v1.1872.00

Pricing

  • Upstage$0.10 / $0.40

Input / output per 1M tokens

AA metrics

Intelligence
—
Speed tok/s
78.69
Blended $/M
—

Arena Elo

  • No Arena snapshot linked.