Solar Pro 4: benchmark results
Upstage's Solar Pro 4 (August 10, 2026), the successor to Solar Pro 3, a cost-efficient flagship reasoning model for agentic office, document and coding workflows with a 524,288-token context. Provider: Upstage. Released 2026-08-10. Access: API.
Aggregate Rating 1688 ± 13, rank #224 of 1635 rated models, from 241 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - jaster (2-shot) - JSICK | 86 | Exact match (%) | 97.2 |
| AA-Omniscience Hallucination Rate | 24.4 | Hallucination Rate (%) | 96.5 |
| Nejumi 4 - BFCL - Live AST | 75 | Accuracy (%) | 95 |
| Nejumi 4 - ALT - Toxicity | 85.74 | Score (%) | 92.5 |
| Nejumi 4 - ALT Average | 89.57 | Score (%) | 92.5 |
| Nejumi 4 - MT-Bench (Japanese) - Roleplay | 99 | Judge rating (1-10, x10) | 91.8 |
| Nejumi 4 - jaster (0-shot) - MMLU-ProX (Japanese) | 90 | Exact match (%) | 91.8 |
| Nejumi 4 - MT-Bench (Japanese) - Coding | 100 | Judge rating (1-10, x10) | 91.5 |
| Nejumi 4 - MT-Bench (Japanese) - STEM | 100 | Judge rating (1-10, x10) | 90.4 |
| Nejumi 4 - jaster (0-shot) - JSQuAD | 81.19 | Character F1 (x100) | 90.1 |
| Nejumi 4 - GLP - Mathematical Reasoning | 96.67 | Score (%) | 89.5 |
| Nejumi 4 - MT-Bench (Japanese) - Math | 100 | Judge rating (1-10, x10) | 89.4 |
Interactive version: theaggregate.ai/model?slug=solar-pro-4 · How It Works · Data refreshed daily, snapshot 2026-10-15.