Neura Intelligence Index
Category Comparison
Benchmark-by-Benchmark
reasoning
| Benchmark | GPT-5.5 | Step3-VL-10B |
|---|---|---|
| GPQA Diamond | 93.6 | — |
| Humanity's Last Exam | 52.2 | — |
| ARC-AGI v2 | 85 | — |
coding
| Benchmark | GPT-5.5 | Step3-VL-10B |
|---|---|---|
| SWE-Bench Pro | 58.6 | — |
agent
| Benchmark | GPT-5.5 | Step3-VL-10B |
|---|---|---|
| BrowseComp | 84.4 | — |
| Toolathlon | 55.6 | — |
| MCP Atlas | 75.3 | — |
math
| Benchmark | GPT-5.5 | Step3-VL-10B |
|---|---|---|
| FrontierMath | 35.4 | — |
| AIME 2025 | — | 87.7 |
general
| Benchmark | GPT-5.5 | Step3-VL-10B |
|---|---|---|
| MRCR v2 | 74 | — |
Category Winners
coding
52.3
math
S
Step3-VL-10B65.9
reasoning
89.7
general
80.8
multimodal
86.6
agent
68.6