OpenAI Just Cut Luna 80%. AI Agents Hit Front-of-Funnel.
On 30 July 2026, OpenAI cut the price of GPT-5.6 Luna by 80% and Terra by 20%. That single line matters more to growth leaders than any model benchmark, because it moves real-time conversational AI across the price threshold where it stops being an internal efficiency tool and starts being a viable customer-facing discovery channel.
The proof is already live in retail. A 24/7 multilingual shopping agent served 30,000 shoppers in two weeks with 92% positive responses. That is not a pilot metric. That is a front-of-funnel surface answering, recommending and converting in the customer's own language.
What actually happened
OpenAI passed on efficiency gains from GPT-5.6 directly to API pricing. Luna, its fastest and cheapest model, now costs 80% less. Terra, the balanced everyday model, costs 20% less. Both cuts also apply to how usage is counted inside Codex and ChatGPT Work.
Crucially, Luna is not a stripped-down chatbot. It uses tools and completes multi-step workflows, which is exactly what a shopping agent needs to check stock, compare specifications and hand off to checkout. OpenAI describes it as delivering performance comparable to models that were frontier-class a year ago, at roughly six cents on the dollar per task.
Why the price cut changes the maths
Front-of-funnel discovery is a high-volume, low-margin game. Every product question, every language, every browsing session is a call. At old prices, running a live conversational agent for anonymous shoppers was hard to justify. At Luna's new price, the unit economics finally clear.
| Model | Price change (30 Jul 2026) | Best fit for growth teams |
|---|---|---|
| GPT-5.6 Luna | 80% cheaper | High-volume customer-facing discovery, support, product Q&A |
| GPT-5.6 Terra | 20% cheaper | Everyday scoped tasks, workspace Q&A where latency matters |
| GPT-5.6 Sol | Fast mode added | Complex reasoning where response time is consequential |
Table: The GPT-5.6 pricing shift. Source: OpenAI.
What the retail case really shows
The 30,000-shopper deployment is the tell. From my observation, most brands still treat AI as a cost centre, aimed inward at content and ops. This case flips it outward. A multilingual agent that greets, recommends and converts is a discovery surface, and in my opinion it behaves less like support and more like a top-of-funnel channel.
I think the multilingual point is underrated. A single agent can serve every market in native language without headcount per region. That collapses one of the oldest barriers to international discovery.
What to do about it
- Map one high-volume customer question your team fields daily. That is your first agent use case.
- Route it through Luna, and reserve the more capable models only for steps where extra intelligence changes the outcome.
- Instrument it. If a live agent becomes a discovery channel, it needs the same attribution rigour as any other. Our guide to making AI traffic countable in GA4 is the place to start.
Cheaper models are becoming a pattern, not a moment. It follows the same logic as Anthropic shipping Opus 5 at Opus 4 prices. The threshold has moved. The question for growth leaders is no longer whether you can afford a conversational agent, but which part of your funnel it belongs in.
Tags