Prime Intellect: INTELLECT-3

INTELLECT-3 is a cutting-edge 106B-parameter Mixture-of-Experts (MoE) model, with 12B active parameters, meticulously post-trained from GLM-4.5-Air-Base. This advanced model leverages supervised fine-tuning (SFT) followed by large-scale reinforcement learning (RL) to achieve exceptional capabilities. It consistently delivers state-of-the-art performance for its size across critical domains such as mathematics, code generation, scientific problem-solving, and general reasoning, frequently surpassing the benchmarks set by many larger frontier models. Designed with a focus on robust multi-step problem solving, INTELLECT-3 maintains high accuracy on complex, structured tasks. Its innovative MoE architecture ensures remarkable efficiency during inference, making it a powerful yet cost-effective solution. With a generous context window of 131K tokens and a maximum output of 4K tokens, it's ideal for extensive conversational applications. Pricing is competitive at $0.20 per 1M input tokens and $1.10 per 1M output tokens. Available on our PRO Access Tier, INTELLECT-3 supports functions, code generation, and streaming capabilities.