OpenAI: GPT-5.6 Terra
OpenAI's flagship model. Long-context specialist with 1.1M window.
Scores by test
Methodology →What you need to know
GPT-5.6 Terra is a high-reasoning model optimized for complex logic and long-form data processing. Its primary differentiator is near-perfect mathematical and strategic performance, evidenced by a 99.7% score on AIME 2025 and maximum internal ratings in strategic analysis and creative problem solving. With a 1.1M context window and a 5/5 rating for long context handling, it is built for deep analysis of massive datasets.
The model maintains high reliability across most operational tasks, achieving 5/5 in structured output, faithfulness, and agentic planning. While it ranks #7 overall, it shows slight relative weakness in classification, constrained rewriting, and tool calling, where it scores 4/5. These are not failures, but they represent the only areas where the model deviates from its otherwise maximum performance profile.
At a blended cost of $11.88/MTok, this is an expensive model. The pricing reflects its position as a premium reasoning engine rather than a general-purpose utility. Developers are paying a significant premium for the precision and the massive context window, making it inefficient for simple chat or basic classification tasks.
Use this model if your application requires complex agentic planning, high-stakes mathematical reasoning, or the processing of million-token documents. Skip this model if you are running high-volume, low-complexity tasks where the high input and output costs would outweigh the marginal gains in intelligence.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models