Best AI models for long arithmetic
Ranked by accuracy on the meo “long arithmetic” domain — one of eleven un-leaked, contamination-resistant domains. Per-domain strength reveals where a model genuinely reasons versus where it pattern-matches; the leaders differ sharply by domain.
Ranking 22 models across the meo-tested roster · as of 2026-06-08.
| # | Model | Lab | accuracy |
|---|---|---|---|
| 1 | OpenAI: GPT-5.5 | openai | 100% |
| 2 | Anthropic: Claude Opus 4.8 | anthropic | 100% |
| 3 | Google: Gemini 3.1 Pro Preview | 53% | |
| 4 | xAI: Grok 4.3 | x-ai | 47% |
| 5 | DeepSeek: DeepSeek V4 Flash | deepseek | 33% |
| 6 | inclusionAI: Ring-2.6-1T | inclusionai | 33% |
| 7 | MoonshotAI: Kimi K2.6 | moonshotai | 33% |
| 8 | Owl Alpha | openrouter | 33% |
| 9 | Perceptron: Perceptron Mk1 | perceptron | 33% |
| 10 | Google: Gemini 3.5 Flash | 27% | |
| 11 | Qwen: Qwen3.7 Max | qwen | 27% |
| 12 | Xiaomi: MiMo-V2.5 | xiaomi | 20% |
| 13 | Qwen: Qwen3.7 Plus | qwen | 20% |
| 14 | Xiaomi: MiMo-V2.5-Pro | xiaomi | 20% |
| 15 | DeepSeek: DeepSeek V4 Pro | deepseek | 13% |
| 16 | StepFun: Step 3.7 Flash | stepfun | 13% |
| 17 | NVIDIA: Nemotron 3 Ultra | nvidia | 7% |
| 18 | MiniMax: MiniMax M3 | minimax | 7% |
| 19 | Google: Gemini 3.1 Flash Lite | 7% | |
| 20 | Tencent: Hy3 preview | tencent | 7% |
| 21 | Arcee AI: Trinity Large Thinking | arcee-ai | 7% |
| 22 | Z.ai: GLM 5.1 | z-ai | 0% |