- MMLU PRO (0-shot, CoT) 81.2
- GPQA-Diamond 71.9
- HLE 11.9
- HMMT'26 65.9
- AIME'26 83
- LCB.v1-6 (cumulative) 77.7
- Tau2-all 72.3
- Terminal Bench 2 10.1
- IFBench 73.1
- Arena-hard-v2 74.7
- Ko-Arena-hard-v2 78.2
Solar Pro 3 (March 2026 update)
Solar Pro 3 (March 2026 update).
11
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
Upstage Mar 2026 Solar Pro 3 (March 2026 update)
Closed weights Flagship
136.1
#220 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- IFBench 73.1% #29 /80
- tau2-bench 72.3% #54 /79
- MMLU-Pro 81.2% #68 /167
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 | |
|---|---|---|---|---|
| Arena-Hard v2.0 | 74.7% | #8 of 24 | Lab model card | |
| IFBench | 73.1% | #29 of 80 | Lab model card | |
| MMLU-Pro | 81.2% | #68 of 167 | Lab model card | |
| Humanity's Last Exam | 11.9% | #131 of 154 | Lab model card | |
| Ko-Arena-Hard v2 (Upstage) | 78.2% | #2 of 2 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 71.9% | #167 of 302 | Lab model card | |
| AIME 2026 | 83% | #32 of 38 | Lab model card | |
| HMMT 2026 | 65.9% | #19 of 19 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 77.7% | #38 of 90 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau2-bench | 72.3% | #54 of 79 | Lab model card | |
| Terminal-Bench 2.0 | 10.1% | #63 of 63 | Lab model card |
Its lab's numbers put it at 136.1, number 220 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 132.7 to 139.6. How we compute it.
- AIME 2026 83
- MMLU-Pro 81.2
- LiveCodeBench v6 77.7
- Arena-Hard v2.0 74.7
- IFBench 73.1
- tau2-bench 72.3
- GPQA Diamond 71.9
- HMMT 2026 65.9
- Humanity's Last Exam 11.9
- Terminal-Bench 2.0 10.1
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.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| Ko-Arena-Hard v2 (Upstage) | 78.2% Ko-Arena-hard-v2 | 78.2% Leaderboard | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.