GPT-6: Does SOL and LUNA's price advantage outweigh their capabilities? Sol: $2 input / $10 output Luna: $0.10 input / $0.50 output Artificial Analysis ran the numbers. Using the same Intelligence Index, running a task on Sol costs about $1.06, while Luna only costs $0.07. This isn’t just about being a bit cheaper—tasks that were previously too expensive to let AI handle continuously can now be thrown into the background to run non-stop. And this time, it’s not just about trading IQ for price. Sol achieves 68.8% on DeepSWE, only 1.1 percentage points behind Claude Fable 5’s 69.9%, but at roughly 80% lower task cost. On AutomationBench, it scores 33.2%, even surpassing Opus 5 max’s 26.9%, while costing only about 1/11th as much. For coding "lie rate," GPT-5.6 Sol was at 10.4%, but GPT-6 Sol slashed it down to 1.3%. So, GPT-6 isn’t like Astra, where it’s just about raising the capability ceiling again. This time, it’s taken Astra’s generation of workflows, agents, and alignment capabilities and slashed them hard under the price curve.