OpenAI launches GPT-5.6 model family with new cache pricing and coding performance gains
OpenAI introduces the GPT-5.6 model family featuring enhanced coding performance, improved efficiency, and a new cache-based pricing structure.
1. Performance and Benchmarking
OpenAI has released the GPT-5.6 model family, consisting of Sol, Terra, and Luna. According to Artificial Analysis, the flagship model, GPT-5.6 Sol, ranks second in the Intelligence Index with a score of 59, trailing Claude Fable 5 by one point. Despite this, Sol is noted for being approximately one-third the cost of its competitor. In the Coding Agent Index, GPT-5.6 Sol leads all evaluations, including DeepSWE, Terminal-Bench v2, and SWE-Atlas-QnA. Additionally, the model family demonstrates high efficiency, using fewer output tokens per task than many comparable models. In the AA-Briefcase benchmark, which evaluates knowledge work, GPT-5.6 Sol achieved the highest Presentation Elo for its visual output quality, though it remains behind Claude Fable 5 in overall rubric and analytical scores.
2. Pricing and Cache Structure
The GPT-5.6 series introduces cache-write pricing to OpenAI’s offerings for the first time. The models are priced at $5/$30, $2.5/$15, and $1/$6 per million input/output tokens for Sol, Terra, and Luna, respectively. OpenAI has maintained a 90% discount for cache reads but implemented a cost premium for cache writes, set at 1.25 times the price of standard input tokens. This change is intended to reflect the memory resources required to store committed input tokens. Furthermore, the models incorporate a "max reasoning effort" setting, allowing users to adjust performance levels to balance intelligence against cost per task.
3. Model Efficiency and Positioning
The GPT-5.6 family establishes a new Pareto frontier for intelligence versus cost. Data indicates that each model in the series outperforms the previous GPT-5.5 generation in efficiency. Within the new lineup, Sol and Luna consistently outperform Terra in terms of intelligence-to-cost ratios. While Sol provides the highest intelligence, the tiered structure of the three models allows users to select options based on specific budgetary and performance requirements. The models are designed to be competitive with other industry leaders like Gemini 3.5 Flash and GLM-5.2, offering lower costs while maintaining or exceeding intelligence benchmarks.