Due to the massive price cut of Luna today, I've created a new type of Pareto graph that I haven't seen before. A combined planning/implementation set.
Since AI keeps recommending using different models for plan/execute, I asked it to combine them into a single pareto frontier.
| Planning → execution | Quality proxy | Estimated credits |
|---|---|---|
| Luna Max → Luna Max | 55.18 | 9.81 |
| Sol Medium → Luna Max | 58.40 | 26.54 |
| Sol High → Luna Max | 60.54 | 33.73 |
| Sol High → Sol High | 61.70 | 129.38 |
| Sol High → Sol Max | 63.73 | 202.88 |
These calculations use Artificial Analysis’s Codex-harness results: DeepSWE and Terminal-Bench for execution, and SWE-Atlas plus its Intelligence Index as planning proxies. Luna costs were reduced by 80% to reflect OpenAI’s July 30 pricing change.