Neura Intelligence Index
Category Comparison
Benchmark-by-Benchmark
reasoning
| Benchmark | Claude Opus 4.5 | DeepSeek-V4-Pro-Max |
|---|---|---|
| GPQA Diamond | 87 | 90.1 |
| ARC-AGI v2 | 37.6 | — |
| Humanity's Last Exam | — | 48.2 |
coding
| Benchmark | Claude Opus 4.5 | DeepSeek-V4-Pro-Max |
|---|---|---|
| SWE-bench Verified | 80.9 | 80.6 |
| SWE-Bench Pro | — | 55.4 |
agent
| Benchmark | Claude Opus 4.5 | DeepSeek-V4-Pro-Max |
|---|---|---|
| OSWorld | 66.3 | — |
| MCP Atlas | 62.3 | 73.6 |
| BrowseComp | — | 83.4 |
| Toolathlon | — | 51.8 |
Category Winners
coding
82.5
reasoning
85.5
general
78.0
agent
67.2