Neura Intelligence Index
Category Comparison
Benchmark-by-Benchmark
reasoning
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Speciale |
|---|---|---|
| GPQA Diamond | 75.4 | — |
| Humanity's Last Exam | — | 30.6 |
math
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Speciale |
|---|---|---|
| AIME 2025 | 70.5 | 96 |
coding
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Speciale |
|---|---|---|
| SWE-bench Verified | 72.7 | 73.1 |
| TerminalBench | 35.5 | — |
multimodal
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Speciale |
|---|---|---|
| MMMU | 74.4 | — |
general
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Speciale |
|---|---|---|
| MMMLU | 86.5 | — |
agent
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Speciale |
|---|---|---|
| TAU-Bench Retail | 80.5 | — |
| Toolathlon | — | 35.2 |
Category Winners
coding
68.0
math
79.0
reasoning
66.7
general
62.9
multimodal
68.6
agent
83.4