OpenAI has turned the latest AI model race into a cost-per-task race. On September 22, the company launched GPT-6 Sol and GPT-6 Luna, two lower-cost models that bring capabilities from GPT-6 Astra to everyday professional work, coding, automation, and high-volume tasks. OpenAI says the models cut API prices by 50% versus the promotional pricing for their GPT-5.6 predecessors: Sol is $2 per million input tokens and $10 per million output tokens, while Luna is $0.10 input and $0.50 output. (OpenAI’s launch announcement)

For marketing leaders, this is more than another model release. Lower inference costs make it economically realistic to run more research, personalization, creative testing, reporting, and agentic workflow steps. The competitive question is shifting from “Can AI do this?” to “How many useful tasks can we afford to delegate—and how will we measure the result?”

The important news is the price-performance curve

OpenAI positions GPT-6 Sol for difficult work tasks and GPT-6 Luna for fast, high-volume jobs such as summarization, extraction, and routine responses. Both are available in the API, and both are rolling out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users; Luna is also available to Free and Go users in the desktop app. (OpenAI availability details)

The difference is not simply that tokens are cheaper. OpenAI says improved inference and prompt caching reduce the cost of repeatedly processing context, with cached input reads discounted by 90%. That matters for marketing systems that reuse brand guidelines, product catalogs, campaign history, audience rules, or a long research brief across many calls. (OpenAI API pricing)

OpenAI’s own AutomationBench figures illustrate the intended shift. In its company-reported test of end-to-end workflows using 47 tools across sales, marketing, operations, support, finance, and HR, GPT-6 Sol at xhigh effort scored 33.2% at a reported $0.27 per task. OpenAI says that was 11.1 times less expensive per task than Claude Opus 5 at max effort, though buyers should treat cross-vendor benchmark claims as directional until they test their own workflows. (OpenAI’s benchmark methodology and results)

Marketing automation can move from pilots to production

At older price points, teams often limited AI agents to high-value experiments: a weekly report, a small batch of ad variants, or a human-reviewed content brief. A lower-cost model tier changes the operating math. An agency can ask Luna to classify thousands of search queries, normalize campaign data, tag creative concepts, or draft structured metadata, then reserve Sol—or a premium model—for judgment-heavy decisions and exception handling.

This is the emerging model-routing pattern: use the least expensive model that meets the quality threshold, escalate ambiguous cases, and log the reason for escalation. For example, a marketing operations agent could use Luna to find anomalies in paid-search data, Sol to explain likely causes and propose tests, and a human to approve budget changes. That architecture is more defensible than sending every task to the most capable model or allowing an agent to act without review.

The same-day launch of Anthropic Claude Opus 5.5 reinforces the direction of travel. Anthropic says Opus 5.5 costs 40% less to run than Opus 5 and is designed for long-running agents, coding, and professional work. (Anthropic’s Opus 5.5 announcement) The result is a buyer’s market: agencies should compare cost per completed business outcome, not headline model quality or token price in isolation.

What changes for AI search and GEO

Cheaper models will increase the volume of machine-mediated discovery. Businesses should expect more agents to read product pages, compare vendors, summarize reviews, classify claims, and assemble recommendations before a person ever reaches a website. That raises the value of clean, consistent, machine-readable information: product facts, pricing, policies, author credentials, evidence, and up-to-date customer proof.

It also raises the cost of being vague. If agents can cheaply process more sources, unsupported claims and conflicting pages become easier to detect—and easier to exclude. GEO is therefore not just about appearing in an AI answer. It is about maintaining a trustworthy evidence layer that survives extraction, comparison, and repeated summarization across models.

A practical 30-day plan for marketing leaders

  1. Inventory repeatable work. List research, reporting, tagging, content QA, lead qualification, and customer-response tasks that consume volume rather than judgment.
  2. Set a quality floor. Define the acceptable accuracy, citation, latency, and escalation rate for each task before choosing a model.
  3. Pilot routing. Run Luna on high-volume classification and drafting, Sol on synthesis and complex workflow steps, and a human on approvals and exceptions.
  4. Measure unit economics. Track cost per qualified lead, approved asset, resolved ticket, or completed report—not just tokens or API calls.
  5. Strengthen the source layer. Consolidate product facts, proof points, policies, and brand rules so every model receives the same current context.

GPT-6 Sol and Luna do not eliminate the need for strategy; they make weak strategy cheaper to execute at scale—and strong operating systems far more powerful. If your business needs a measurable plan for AI marketing, automation, or GEO, contact Real Internet Sales at 803-708-5514 or visit realinternetsales.com.

Sources: OpenAI: Introducing GPT-6 Sol and Luna; OpenAI API pricing; Reuters: OpenAI expands GPT-6 lineup; TechCrunch: OpenAI launches GPT-6 Sol and Luna; Anthropic: Claude Opus 5.5.