- Terminal-Bench v2.1 (Artificial Analysis score quoted by Upstage) 57
- BrowseComp* (in-house) 49.2
- SWE-Bench Verified (OpenHands)* (in-house) 70.6
- MCP-Atlas* (in-house) 61.4
- APEX-Agents* (in-house) 18.7
- τ³-Banking (Artificial Analysis score quoted by Upstage) 23
- AA-LCR (Artificial Analysis score quoted by Upstage) 71
- GPQA Diamond (Artificial Analysis score quoted by Upstage) 89
- MMLU-Pro* (in-house) 86.3
- AIME 2026* (in-house) 95.3
- KMMLU-Pro* (in-house) 79.2
- KBL (law)* (in-house) 77.5
- KorMedMCQA (medicine)* (in-house) 93.2
- Ko-GDPval* (in-house) 87.3
Solar Pro 4
Solar Pro 4 ranks first on KMMLU-Pro.
30
Benchmarks with a score
1
Ranked first, out of five or more models
1
In the top three
Upstage Aug 2026 Solar Pro 4
Closed weights Flagship
144.8
#135 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- AA-LCR (Artificial Analysis Long Context Reasoning) 71% #7 /38
- APEX-Agents 18.7% #11 /12
- KMMLU-Pro 79.2% #1 /7
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMLU-Pro | 86.3% | #15 of 167 | Lab model card | |
| KMMLU-Pro | 79.2% | #1 of 7 | Lab model card | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 71% | #7 of 38 | Lab model card | |
| StudyArena Text | 1,249 | #25 of 85 | Official leaderboard | |
| LMArena (Elo) | 1,387 | #147 of 335 | Official leaderboard | |
| OpenAI MRCR v2 | 14.4% | #43 of 47 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 89% | #43 of 302 | Lab model card | |
| AIME 2026 | 95.3% | #12 of 38 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 70.6% | #75 of 155 | Lab model card | |
| DesignArena 3D Design | 1,162 | #72 of 136 | Official leaderboard | |
| DesignArena Game Dev | 1,168 | #78 of 143 | Official leaderboard | |
| DesignArena Code | 1,171 | #81 of 143 | Official leaderboard | |
| DesignArena Website | 1,164 | #93 of 151 | Official leaderboard | |
| ALE-Bench | 1,256 | #79 of 126 | Official leaderboard | |
| DesignArena Data Visualization | 1,175 | #75 of 118 | Official leaderboard | |
| DesignArena UI Component | 1,141 | #92 of 139 | Official leaderboard | |
| Code Arena | 1,369 | #95 of 128 | Official leaderboard | |
| DesignArena Agentic Overall | 1,076 | #55 of 74 | Official leaderboard | |
| DesignArena Frontend Web App | 1,087 | #47 of 53 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| τ³-bench Banking | 23% | #13 of 22 | Lab model card | |
| BrowseComp | 49.2% | #64 of 81 | Lab model card | |
| MCP Atlas | 61.4% | #37 of 46 | Lab model card | |
| Terminal-Bench 2.1 | 57% | #47 of 56 | Lab model card | |
| Agent Arena | -22.60% | #56 of 56 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| KBL | 77.5% | #1 of 4 | Lab model card | |
| KorMedMCQA | 93.2% | #1 of 4 | Lab model card | |
| Ko-GDPval | 87.3% | #1 of 2 | Lab model card | |
| FoodTruck Bench | $13,985 | #37 of 59 | Official leaderboard | |
| APEX-Agents | 18.7% | #11 of 12 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 65.3 | #119 of 185 | Official leaderboard |
Its lab's numbers put it at 144.8, number 135 of 404
The Self-Reported Index fits one score to the 10 capability results Upstage reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 142.6 to 147.0. How we compute it.
- AIME 2026 95.3
- GPQA Diamond 89
- MMLU-Pro 86.3
- AA-LCR (Artificial Analysis Long Context Reasoning) 71
- SWE-bench Verified 70.6
- MCP Atlas 61.4
- Terminal-Bench 2.1 57
- BrowseComp 49.2
- τ³-bench Banking 23
- APEX-Agents 18.7
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.