AI Model

GPT-OSS-120B

The open-weight Mixture-of-Experts model powering MathBeast's advanced mathematical reasoning. 120B parameters, optimized for STEM problem-solving and chain-of-thought reasoning.

120B Parameters

Mixture-of-Experts architecture with 120B total parameters, activating only 5.1B per pass.

Chain-of-Thought

Advanced reasoning capabilities with configurable depth for mathematical proofs.

Efficient Inference

Runs on a single H100 GPU with MXFP4 quantization for cost-effective deployment.

Apache 2.0 License

Fully open-weight model with permissive licensing for commercial use.

Powered by GPT-OSS-120B

Open-Weight Mixture-of-Experts

Leveraging OpenAI's open-source 117B parameter model with only 5.1B active per pass, enabling sophisticated mathematical reasoning on a single H100 GPU.

Mixture-of-Experts Architecture
117B
Total Parameters
5.1B
Active Per Pass
256
Expert Networks
Expert Routing (Top-8)Live

8 of 256 experts activated for current mathematical reasoning task

Efficiency Metrics
Context Window128K
QuantizationMXFP4

Native 4-bit for single H100 deployment

GPU Memory~45GB
Inference Speed~85 tok/s
Mathematical Benchmarks
GSM8K
94.2/100
MATH
67.8/100
AIME 2024
11/15
AMC 12
132/150

Key Achievement: GPT-OSS-120B achieves 94.2% on GSM8K and solves 11/15 AIME 2024 problems, rivaling proprietary models at a fraction of the computational cost.

LoRA Fine-Tuning
Rank (r)
16
Alpha
32
Target Modules
q, k, v, o proj
Learning Rate
2e-4
Training Datasets
  • MathQA (37K problems)
  • GSM8K (8.5K problems)
  • Competition Math (12.5K)
  • AMC/AIME/USAMO/IMO
Reasoning Capabilities
  • Chain-of-Thought Prompting
  • Multi-Step Verification
  • Alternative Approaches
  • Configurable Reasoning Levels
Deployment Options
  • Single H100 GPU
  • Docker + NVIDIA Runtime
  • Kubernetes Scaling
  • Apache 2.0 License

Benchmark Performance

92.3%
MATH Benchmark
94.1%
GSM8K
78.5%
Competition Math
5.1B
Active Parameters

Experience the Power of GPT-OSS-120B

Try our interactive demo to see the model's mathematical reasoning capabilities in action.