OpenAI's newest model, GPT-6 Luna, costs $0.10 per million input tokens and $0.50 per million output tokens, according to the company's published API pricing page updated October 29, 2026. That is a 300x price spread from GPT-5.5 Pro, which runs $30 per million input tokens and $180 per million output tokens. The gap is the widest OpenAI has ever shipped between its cheapest and most expensive production models.
GPT-6 Luna also carries a 1.05 million token context window and offers cached input at $0.01 per million tokens. Those specs put it closer to mid-tier models like GPT-5.4, which costs $2.50 per million input tokens, than to the nano tier that previously held the floor. OpenAI now has two Luna-branded models: GPT-5.6 Luna at $0.20 per million input tokens, and GPT-6 Luna at half that price.
The pricing matters because it lands at a moment when developers are building agents that make thousands of calls per task. At $0.10 per million input tokens, a developer can run a 10,000-token prompt 100,000 times for $100. The same volume on GPT-5.5 Pro costs $30,000. That difference is not a rounding error; it changes which use cases are economically viable.
It's not clear this will hold. OpenAI has revised pricing before, and the GPT-5.1 Codex line was deprecated within a year of launch. Developers who build on Luna's price point today may face a different bill in 2027. Cached input pricing at $0.01 per million tokens also assumes high cache hit rates, which most workloads don't achieve.
The broader pattern is a price war at the bottom of the stack. Google's Gemini Flash models and Anthropic's Haiku tier have been pushing input prices below $0.50 per million tokens for the past year. OpenAI's move to $0.10 for a model with a 1.05 million token context window is a direct response. The question is whether the quality at that price point holds up for real workloads, and the pricing page doesn't answer that.
What GPT-6 Luna Actually Costs in Practice
A developer running a typical agent loop makes roughly 50 calls per task, each averaging 2,000 input tokens and 500 output tokens. On GPT-6 Luna, that task costs about $0.0125 in input and $0.0125 in output, or roughly 2.5 cents total. On GPT-5.4, the same task costs about $0.25 in input and $0.075 in output, or 32.5 cents. The 13x difference per task compounds fast for companies running thousands of tasks per day.
Cached input at $0.01 per million tokens is the real headline for repeat-heavy workloads like RAG pipelines and chatbots with shared system prompts. A 5,000-token system prompt cached across 100,000 calls costs $5 on GPT-6 Luna versus $50 on GPT-5.4. But cache hit rates depend on prompt structure and traffic patterns, and most teams overestimate their hit rates when they first deploy.
The Quality Question Nobody Has Answered Yet
OpenAI's pricing page lists GPT-6 Luna as "Current" status, meaning it is available for production use. It does not include benchmark scores. The company's model card for GPT-6 Luna has not been published as of October 29, 2026, so developers have no independent way to compare its reasoning, coding, or instruction-following performance against GPT-5.4 or GPT-5.5.
BenchLM.ai, which tracks OpenAI API pricing, lists 38 current models on OpenAI's roster. That is more than double the number of models OpenAI offered two years ago. The proliferation creates a selection problem: developers now face a matrix of price, context window, and capability tiers with limited guidance on which model fits which task.
How This Fits the Broader Pricing Trend
Two years ago, GPT-4o launched at $2.50 per million input tokens and was considered cheap. Today, GPT-6 Luna costs 25x less than that. The decline has been steep and consistent, roughly an order of magnitude per year for the cheapest tier. GPT-4.1 nano at $0.10 per million input tokens was the previous floor, and it launched with a 1 million token context window. GPT-6 Luna matches that context window at the same price but with a newer architecture.
The spread between cheapest and most expensive has also grown. GPT-5.5 Pro at $30 per million input tokens is 300x more expensive than GPT-6 Luna. That spread reflects a market segmentation strategy: OpenAI is pricing for both high-volume, low-margin agent workloads and premium, low-volume reasoning tasks. Whether that segmentation holds depends on whether customers actually need the Pro tier for tasks that Luna can't handle.
The Developer Trade-Off: Cost vs. Capability
The real decision for engineering teams is not which model is cheapest. It is which model is cheapest for a given task. A team running a customer support chatbot with simple intent classification and templated responses may find GPT-6 Luna indistinguishable from GPT-5.4 in output quality. A team running code review or legal document analysis will likely find the gap much wider.
OpenAI's own product lineup hints at this segmentation. The company offers GPT-5.5 Pro at $30 per million input tokens for tasks that need deep reasoning, and GPT-6 Luna at $0.10 for tasks that don't. The 300x spread is not a bug; it is the business model. OpenAI wants to capture both the high-volume, price-sensitive agent market and the low-volume, premium reasoning market.
The risk for developers is lock-in. Once a team builds a pipeline around GPT-6 Luna's latency and output characteristics, switching to a competitor's model requires re-testing and re-tuning. OpenAI knows this, which is why the pricing is aggressive now. The moat is not the model; it is the integration cost of leaving.
For more on how AI pricing is reshaping developer tooling, see our AI coverage.
What Developers Should Watch Next
The missing piece is independent evaluation. Until someone publishes benchmark results for GPT-6 Luna on standard tasks like MMLU, HumanEval, and GSM8K, the $0.10 price tag is just a number. Developers should treat it as a candidate for high-volume, low-stakes workloads like classification, summarization, and simple extraction, not as a drop-in replacement for GPT-5.4 in production systems.
OpenAI's next pricing update is expected in November 2026, when the company typically refreshes its model lineup ahead of the holiday quarter. If GPT-6 Luna's price holds through that update, it signals a genuine strategic bet on volume over margin. If it rises, the $0.10 figure was likely an introductory rate designed to drive adoption.
For now, the math is simple. A team spending $10,000 per month on GPT-5.4 input tokens could cut that to $400 by switching to GPT-6 Luna for the same token volume. The real cost is the engineering time to test whether the cheaper model produces acceptable output. That test, not the price, will determine whether Luna becomes the default for agent workloads.
Sources: BenchLM.ai OpenAI API pricing, OpenAI API pricing documentation