Mathematical reasoning and calculation on TPS-CalcBench (test)
90.2KPIgpt-5.2
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| gpt-5.2mode=RAG-EQ (retrieval-augmented result)2026.04 | 90.2 | 2.4 | 38 | |
| gpt-5.2mode=Base (zero-shot baseline)2026.04 | 87.85 | — | — | |
| deepseek-v3.1mode=RAG-EQ (retrieval-augmented result)2026.04 | 84.1 | 4.9 | 44 | |
| aws.claude-opus-4.5mode=RAG-EQ (retrieval-augmented result)2026.04 | 83.6 | 5.8 | 52 | |
| aws.claude-sonnet-4.5mode=RAG-EQ (retrieval-augmented result)2026.04 | 79.4 | 7.6 | 61 | |
| deepseek-v3.1mode=Base (zero-shot baseline)2026.04 | 79.22 | — | — | |
| aws.claude-opus-4.5mode=Base (zero-shot baseline)2026.04 | 77.8 | — | — | |
| aws.claude-sonnet-4.5mode=Base (zero-shot baseline)2026.04 | 71.81 | — | — | |
| gemini-3-flash-previewmode=RAG-EQ (retrieval-augmented result)2026.04 | 57.3 | 15.2 | 71 | |
| gemini-3-flash-previewmode=Base (zero-shot baseline)2026.04 | 42.15 | — | — | |
| MiniMax-M2.5mode=RAG-EQ (retrieval-augmented result)2026.04 | 28.7 | 15.6 | 68 | |
| MiniMax-M2.5mode=Base (zero-shot baseline)2026.04 | 13.12 | — | — |