Upstage Solar Open 2 Card - HRM8K: leaderboard
HRM8K bilingual Korean-English math reasoning benchmark, run by Upstage. Upstage Solar Open 2 model card, Korean benchmark table (July 2026); accuracy in percent, higher is better; DeepSeek-V4-Flash at max effort.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash (Max) | 93.4 | #155 (DeepSeek V4 Flash) |
| 2 | Solar Open2 250B | 92.2 | #173 |
| 3 | GPT-5.4 Mini | 91.3 | #228 |
| 4 | MiMo-V2.5 | 90.7 | #196 |
| 5 | Claude Haiku 4.5 | 90.6 | #287 |
| 6 | Solar Open 100B | 87.6 | #415 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=upstage-solar-open-2-card-hrm8k · How It Works · Data refreshed daily, snapshot 2026-11-05.