Key Takeaways
- OpenAI accidentally revealed three GPT-5.6 Pro variants: Luna Pro, Terra Pro, and Sol Pro
- Sol Pro hit 31.5% pass rate on genomics benchmark, beating Claude Opus 4.8 at 16%
- Pro boosts shrink at higher tiers: Luna Pro gains 7 points over standard, Sol Pro gains only 2.8
OpenAI's Pro tier is about to get complicated. A benchmark paper on genomics, published by the company, lists three separate Pro models for GPT-5.6: Luna Pro, Terra Pro, and Sol Pro. This is the first time OpenAI has shown Pro as anything other than a single flagship product.
The leak appears accidental. OpenAI announced the standard GPT-5.6 lineup in late June, splitting it into Sol for the hardest tasks, Terra for high-volume business workloads, and Luna for faster everyday queries. Pro variants weren't mentioned. Now they've shown up in a results table the authors probably didn't expect anyone to scrutinize this closely.
What the benchmark numbers show
The genomics benchmark measures pass rate, meaning how often a model completes a multi-step analysis correctly. Sol Pro scored 31.5%, the highest of all 60 tested models. Standard Sol hit 28.7%. For comparison, Claude Opus 4.8 landed at 16%.
The interesting pattern is how Pro gains shrink as you move up the model ladder:
| Model | Standard (max) | Pro (Extended) | Gap |
|---|---|---|---|
| GPT-5.6 Luna | 16.5% | 23.6% | +7.1 points |
| GPT-5.6 Terra | 23.3% | 28.5% | +5.2 points |
| GPT-5.6 Sol | 28.7% | 31.5% | +2.8 points |
Luna Pro gains a full seven percentage points over standard Luna. Sol Pro gains less than three. Extra compute lifts weaker models more, which makes sense. The ceiling gets harder to push against.
One striking data point: Terra Pro at 28.5% nearly matches standard Sol at 28.7%. That means the high-volume Pro variant performs almost as well as the best standard flagship. For teams running heavy workloads, that could matter.
Why this breaks from OpenAI's Pro strategy
ChatGPT Pro launched as a simple proposition. Pay $200 per month, get the single best model OpenAI offers. No choices, no tiers within the tier. Just the top.
A three-model Pro lineup changes that logic. Users would pick between speed (Luna Pro), throughput (Terra Pro), and maximum reasoning power (Sol Pro) based on the task. It's the same structure Anthropic uses with Haiku, Sonnet, and Opus, but applied to the premium tier rather than the standard one.
This isn't confirmed for ChatGPT. The model names appear only in a benchmark table, not an official announcement. OpenAI could be running these variants internally without plans to ship them to subscribers. But the naming convention mirrors the standard lineup too closely to be coincidence.
What OpenAI isn't saying
The paper reports token usage for standard GPT-5.6 models as a proxy for compute cost. Sol at its highest setting uses about 33,200 tokens per run. For the Pro variants, that number is missing.
The authors say no comparable token accounting was available. The more likely explanation: OpenAI doesn't want to reveal how much compute these Pro runs actually consume. That figure would let competitors and customers estimate real costs, which OpenAI has consistently kept opaque.
What this means for teams building on OpenAI
If this lineup ships, API users would gain more granular control over the capability-latency tradeoff. A workflow could route simple queries to Luna Pro for speed, bulk processing to Terra Pro for efficiency, and only the hardest reasoning tasks to Sol Pro.
Pricing is the open question. Does Pro become one subscription with access to all three variants? Do the variants carry different rate limits? The paper offers no hints. Teams using tools like Zapier or Make to orchestrate AI workflows will want clarity before rebuilding their routing logic.
Disclosure
Some links in this post are affiliate links — Logicity earns a commission if you sign up, at no extra cost to you. We only link products we have used or actively recommend.
Logicity's Take
The shrinking Pro gains at higher tiers suggest diminishing returns on raw compute. Luna Pro's seven-point jump versus Sol Pro's three-point bump implies that model architecture matters more than extra inference time once you reach flagship scale. For product teams, the real decision isn't 'use Pro or not' — it's 'which Pro variant matches this task's latency budget.' The absence of token data for Pro runs is telling. OpenAI likely burns significantly more compute than the standard tiers, and revealing that would undercut the $200/month value proposition. Expect pricing complexity if these variants ship publicly.
Frequently Asked Questions
What are the three GPT-5.6 Pro models?
Luna Pro, Terra Pro, and Sol Pro. They mirror the standard GPT-5.6 lineup but with extended compute for higher performance.
How does Sol Pro compare to Claude Opus 4.8?
Sol Pro hit 31.5% pass rate on the genomics benchmark. Claude Opus 4.8 scored 16%, roughly half.
Will these Pro variants be available in ChatGPT?
Unknown. The models appear only in a benchmark paper, not an official product announcement.
How much does ChatGPT Pro cost?
The current ChatGPT Pro subscription is $200 per month. Pricing for a multi-model Pro tier hasn't been announced.
Another recent story on AI model policy changes from major labs
Need Help Implementing This?
Building AI workflows that route between model tiers? Our team can help you design efficient pipelines. Contact us at hello@logicity.in.
Source: The Decoder / Maximilian Schreiner
Huma Shazia
Senior AI & Tech Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
