OpenAI unveils three-tier Pro models for ChatGPT in benchmark paper

Key takeaway

• OpenAI's new benchmark paper reveals it is testing three Pro variants of its GPT-5.6 models—a first for ChatGPT Pro, which has historically been a single top-tier offering.

• The strongest variant, Sol Pro, achieved a 31.5 percent pass rate on the genomics benchmark, outperforming both its standard counterpart and Claude Opus 4.8.

• This shift would let users select Pro models by speed, throughput, or reasoning power rather than buying a single expensive tier, though it is unclear whether OpenAI will actually release them as a product.

3 Key Points

FAQ

How do the Pro models compare to their standard versions?

Luna Pro gains seven percentage points over standard Luna, while Terra Pro reaches 28.5 percent (nearly matching standard Sol at 28.7 percent). Sol Pro gains less than three points over standard Sol, showing that the Pro boost is larger for weaker tiers.

Will these Pro models actually be released in ChatGPT?

It is unclear. The three Pro variants appear only in the benchmark results table; the paper does not confirm they will ship as a product in ChatGPT.