Solar Mini 4: benchmark results
Upstage's compact agent-oriented MoE (35B total, 3B active) with a 512K context, offered through the Upstage Console API as the successor to Solar Mini. Provider: Upstage. Released 2026-09-22. Access: API.
Aggregate Rating 1676 ± 19, rank #445 of 10556 rated models, from 75 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA Long Context Reasoning | 83.33 | Accuracy (%) | 95.9 |
| AA-Omniscience Hallucination Rate | 35.83 | Hallucination Rate (%) | 88.4 |
| AA Humanity's Last Exam | 25.76 | Accuracy (%) | 70.6 |
| LLM Stats Score | 32.28 | LLM Stats Score (conservative rating) | 66.5 |
| AA-Omniscience v1.1 Hallucination Rate | 36.53 | Hallucination Rate (%) | 65.7 |
| AA Omniscience | -10.82 | Score | 63 |
| AA CritPt | 1.43 | Accuracy (%) | 58.6 |
| AIIQ Composite IQ | 107 | Composite IQ (imported) | 48.6 |
| AA GDPval | 1092.1 | ELO | 47.5 |
| AA-Briefcase - PDF - Rubric Pass Rate | 31.03 | Rubric Pass Rate (%) | 44.4 |
| AA SciCode | 47.57 | Accuracy (%) | 40.7 |
| AA-Omniscience Index - Law | -6.44 | Omniscience Index | 40.6 |
Interactive version: theaggregate.ai/model?slug=solar-mini-4 · How It Works · Data refreshed daily, snapshot 2026-11-17.