An OpenAI paper has revealed three new GPT-5.6 Pro models for the first time, including Luna Pro, Terra Pro, and Sol Pro. This marks a significant departure from the previous single-top-tier model structure for the Pro tier. The paper suggests that users may soon have the option to choose between models optimized for speed, throughput, and maximum reasoning power, depending on their specific needs. The details come from a genomics benchmark paper, which includes results for these Pro variants labeled as 'Pro (Extended)' runs. Source: thedecoder
The benchmark results show that Sol Pro achieves a pass rate of 31.5 percent, outperforming the standard Sol model at 28.7 percent and surpassing the best non-GPT model, Claude Opus 4.8, at 16.0 percent. The pass rate measures how often a model completes a multi-step analysis without errors and arrives at the correct final answer. Comparisons between the standard tiers and their Pro variants highlight the performance gains, with Luna Pro showing the highest improvement at +7.1 points over its standard version. Source: thedecoder
The paper does not confirm whether these models will be available in ChatGPT, and token usage for the Pro variants remains undisclosed. The names of the Pro models are derived from the benchmark table, and the paper notes that no comparable token accounting was available for the Pro runs. However, the omission of token usage data suggests OpenAI may be withholding that information. Source: thedecoder
