OpenAI Paper Leaks New Tiered GPT-5.6 Pro Model Lineup

A recent genomics benchmark paper released by OpenAI has inadvertently pulled back the curtain on the future of ChatGPT Pro. Instead of a single flagship model, the data suggests OpenAI is preparing a sophisticated three-tier "Pro" lineup designed to balance reasoning power, throughput, and cost.

The Three Pillars: Sol, Terra, and Luna Pro

For much of its history, OpenAI’s "Pro" offering has functioned as a single, monolithic tier—the absolute best model available for power users. However, the leaked genomics paper introduces three distinct "Pro (Extended)" variants that mirror the existing GPT-5.6 architecture: Sol Pro, Terra Pro, and Luna Pro.

Based on the benchmarks, these models are categorized by their intended utility:

  • Sol Pro: The high-reasoning powerhouse designed for the most complex, multi-step analytical tasks.
  • Terra Pro: A high-volume variant optimized for massive business workloads and enterprise-scale throughput.
  • Luna Pro: A lightweight, fast, and cost-effective variant meant for high-frequency everyday queries.

Benchmarking the Performance Leap

The genomics benchmark provides concrete evidence of the performance delta between standard models and their Pro counterparts. The study measured a "pass rate"—the ability to complete a multi-step analysis without error—across a 129-task suite.

The results indicate that the Pro variants offer significant intelligence boosts, particularly for the lower tiers. Sol Pro emerged as the undisputed leader, achieving a 31.5% pass rate. This outperforms the standard Sol model (28.7%) and significantly eclipses the industry's current heavyweights, such as Claude Opus 4.8, which scored 16.0%.

Interestingly, the "Pro boost" is not uniform. The data shows that extra compute provides diminishing returns as model complexity increases. While Luna Pro saw a massive 7-point jump over its standard version, Sol Pro saw a more modest increase of less than 3 points. Notably, Terra Pro reached a 28.5% pass rate, meaning OpenAI's high-volume Pro model performs nearly as well as the standard flagship Sol model.

A Strategic Shift in AI Deployment

This revelation marks a fundamental shift in OpenAI’s product strategy. By moving away from a "one-size-fits-all" Pro tier, OpenAI is addressing the primary friction point for developers and enterprises: the trade-off between intelligence and efficiency.

Providing a "Terra Pro" that matches flagship intelligence but is optimized for volume allows companies to scale AI agents without the prohibitive costs of a pure reasoning model. Similarly, a "Luna Pro" offers a middle ground for users who need more reliability than a standard model but cannot justify the latency of a full Sol run.

While OpenAI has not officially commented on these names or the specific token costs—which were notably absent from the paper's compute accounting—the technical roadmap is clear. The era of the single flagship is ending, replaced by a specialized ecosystem of professional-grade models.

Key Takeaways

  • Tiered Pro Architecture: OpenAI is likely moving toward a three-model Pro lineup (Sol, Terra, and Luna) rather than a single top-tier subscription.
  • Superior Reasoning: Sol Pro set a new benchmark in genomics with a 31.5% pass rate, significantly outperforming competitors like Claude Opus 4.8.
  • Efficiency Gains: The new structure allows users to choose between maximum reasoning power, high-volume throughput, or cost-effective speed.