OpenAI's GPT-6 Sol and Luna aim at agent work at lower cost

OpenAI introduced GPT-6 Sol and Luna, two additions to the GPT-6 family that keep the company's split naming: the number marks the generation, while Sol, Terra, and Luna are capability tiers that can advance on their own schedule. Both models target everyday work, and the pitch is again about cost. OpenAI says it cut API prices for Sol and Luna by 50 percent compared with GPT-5.6 promotional pricing, and that Luna at high effort improves on its predecessor by 5.4 percentage points while costing 58 percent less per task.

The comparisons OpenAI leads with are against Claude Opus 5, and they are about workflow automation rather than raw reasoning. On AutomationBench, a test of business tasks spread across applications, OpenAI says GPT-6 Sol at its highest effort setting outperforms Opus 5 at max effort while costing about 9 percent as much per task. On Agents' Last Exam, which scores agents on complex professional workflows, it reports Sol at max effort reaching 56.4 percent, above Opus 5's best result there at 60 percent lower cost per task.

Those figures come from OpenAI and its own benchmarks, so treat the head-to-head numbers as vendor claims until independent tests land. The pattern across both labs this week is the same: the frontier fight is shifting from who scores highest to who delivers a given level of agent performance for the least money.

Why it matters

If you build agents, the cost per completed task is now the number to watch, not the top benchmark score. OpenAI is claiming order-of-magnitude cost advantages over a rival flagship on workflow tests, which is worth checking against your own pipeline before you commit a vendor for the next year.