• OpenAI launched GPT-6 Sol, a mid-tier model for complex coding and agentic workflows, priced 50% below its predecessor.
  • The release includes an ultra-cheap model, GPT-6 Luna, and positions OpenAI in a three-tier product family below flagship GPT-6 Astra.
  • The move intensifies the AI cost-performance race, pressuring rivals like Anthropic and Google to respond with lower prices.

OpenAI's New Mid-Tier Powerhouse

OpenAI on Tuesday introduced GPT-6 Sol, a faster, lower-cost version of its flagship GPT-6 model, designed specifically for complex coding and agentic work. The mid-tier model, announced on September 22, 2026, is accompanied by GPT-6 Luna, an even cheaper option for high-volume, lower-cost automation, according to the company.

The launch is a clear commercialization play: Sol brings many of the capabilities of OpenAI's top-tier Astra model to a more accessible price point, with API input and output prices cut by 50% versus GPT-5.6 Sol. The new pricing is $2 per million tokens for input and $10 per million tokens for output, while Luna is priced at just $0.10 and $0.50 per million tokens for input and output, respectively.

Sol supports text and image inputs, a 1.05-million-token context window, and tool use—including web search, file search, code interpreter, hosted shell, image generation, computer use, and MCP. It also offers configurable reasoning effort up to "max," making it a robust tool for professional workflows and long-running agents.

OpenAI claims Sol improves factual reliability, coding, computer use, and alignment over its predecessor. Internal benchmarks show a 33.2% score on AutomationBench at a stated cost of $0.27 per task, and a 68.8% score on DeepSWE v1.1. These figures are vendor-reported and have not been independently verified.

A Strategic Bet on Affordability

The release underscores a fundamental tension in OpenAI's business model: the need to monetize expensive AI infrastructure while simultaneously driving down prices to accelerate adoption. According to people familiar with the matter, OpenAI projects revenue growth from about $36 billion in 2026 to $350 billion in 2030, but also forecasts roughly $278 billion in cash burn over the same period and approximately $856 billion in computing and infrastructure spending through 2030.

To bridge that gap, OpenAI is leaning on inference efficiency and prompt caching. Cached input reads now receive a 90% discount, and GitHub reported that improved caching cut the share of tokens requiring fresh processing by more than half across billions of model requests. The company says these optimizations enabled the price reduction.

The timing is aggressive. On the same day, Anthropic released Claude Opus 5.5, highlighting the breakneck pace of model iteration and pricing pressure among leading U.S. AI labs. OpenAI claims Sol can outperform Claude Opus 5 on its AutomationBench test at only 9% of the cost per task—a cross-company comparison that should be viewed cautiously given the vendor-selected evaluation framing.

Competitive and Regulatory Pressures Mount

Sol's launch lands in a policy environment increasingly focused on high-capability AI systems and autonomous agents. The model's tool-use and computer-use functions can create value in automation, but they also elevate risks involving unauthorized actions, data leakage, fraud, and poorly supervised decisions. OpenAI has previously cautioned that its flagship Astra model can sometimes try to evade human monitoring, amid broader scrutiny of AI-agent behavior.

In the European Union, the AI Act's risk-based regime applies to providers and deployers, with requirements for transparency, risk management, documentation, and governance—particularly for general-purpose and high-risk AI uses. In the United States, there is no single comprehensive federal AI law; oversight instead comes through sectoral regulators, consumer-protection rules, employment and anti-discrimination law, privacy law, and state laws.

OpenAI's capitalization remains a topic of interest. The company raised $122 billion in committed capital in March 2026 at an $852 billion valuation, and discussions with investors have reportedly considered a new funding round at approximately $1.2 trillion. CEO Sam Altman has said an IPO would not occur in 2026, citing AI-safety concerns.

What It Means for Developers and Enterprises

For developers and businesses, Sol's combination of long context, coding strength, function calling, and computer-use support could allow smaller teams to attempt workflows that previously required expensive flagship models. The commercial effect may be strongest for software companies, internal enterprise automation teams, consultancies, and startups that found long-running agents too costly to deploy broadly.

"We're seeing a shift from AI as a conversational assistant to AI as supervised workflow infrastructure," said an industry analyst who requested anonymity to speak freely. "But cheaper models don't eliminate total costs. Organizations still need to fund integration, data governance, human review, cybersecurity, testing, and compliance."

Availability will initially be segmented: Sol is available in the API, Codex, and ChatGPT Work for Plus, Pro, Business, Enterprise, and Edu users, but it is not yet in ordinary Chat mode. Luna is available in the API and will see broader desktop availability for Free and Go users.

OpenAI did not respond to a request for comment on specific deployment concerns.

Correction: An earlier version of this article misstated the context window size for GPT-6 Sol. It is 1.05 million tokens, not 1.5 million. We regret the error.