INTELLECT-MATH
INTELLECT-MATH applies PRIME-RL reinforcement learning to Prime Intellect's supervised Qwen-7B-Math adaptation. The study examines how the starting policy affects mathematical reasoning.
Research release
What the release provides
The model learns first from verified QwQ outputs, then from reinforcement learning. The release includes the completed run and an earlier step-47 checkpoint for comparison.INTELLECT-MATH official model card
How to use it well
Practical guidance. Treat this as a specialist mathematical model. Verify the final calculation with an independent method when you need a reliable answer.
Why the starting policy matters
The study used 733,000 verified question-and-response pairs for supervised training. Prime reports that this initialization reached comparable earlier-system performance with fewer reinforcement-learning steps. That result describes the published experiment, not a general speed guarantee.INTELLECT-MATH research release
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Context window | 4,096 tokens Configured position limit. Serving limits depend on the runtime and settings. INTELLECT-MATH model configuration |
| Base model | Qwen-7B-MathINTELLECT-MATH research release |
| Repository license | MITINTELLECT-MATH official model card |
Related models
- INTELLECT-MATH-SFT
Prime Intellect ยท INTELLECT-MATH
A supervised math-reasoning model that supplies the starting policy for INTELLECT-MATH.
Launch date not confirmedResearch releaseDownloadable