AI / ML3 min read
OpenAI Announces GPT-5 Training Completion with 200B Parameters
Tags AI · Infrastructure
OpenAI Blog·
OpenAI revealed the completion of GPT-5 training, a 200 billion parameter model that achieves 92% on MMLU benchmark, surpassing GPT-4 by 18 percentage points. The model will be available through API starting Q3 2026 with enterprise pricing at $0.0075 per 1K tokens.
Technical significance
GPT-5's 200B parameter scale represents a 6x increase over GPT-4, potentially enabling more complex reasoning and multilingual capabilities. The 92% MMLU score suggests significant advances in general knowledge and factual reasoning, which could accelerate AI adoption across enterprise workflows and reduce the need for specialized fine-tuning.