k0-math
Moonshot's first reasoning model, math-only, claimed to beat o1-mini and o1-preview on four Chinese math exams, two months after o1-preview.
One of the first Chinese long-chain-of-thought reasoning releases following o1-preview. Narrow scope (math); reached about 83% of o1-mini on AIME by Moonshot's own account. Direct precursor to the k1.5 RL recipe.
- Date
- Saturday, 16 November 2024
- Lab
- Moonshot AI
- Kind
- model
- Access
- app only
Figures
| Measure | Value | Measured by |
|---|---|---|
| AIME vs o1-mini | 83% of o1-mini's score, per company | company |
| OMNI-MATH vs o1-mini | 90% of o1-mini's score, per company | company |
Company-claimed benchmarks reported by Global Times; Moonshot said it struggled with geometry and over-thought simple arithmetic.
Sources
This record was checked against its sources on 6 October 2026. How we check