GPT-5.6 price cuts: Luna down 80%, Terra down 20%

OpenAI says GPT-5.6 has moved further along the price-performance frontier: GPT-5.6 Luna is 80% cheaper, GPT-5.6 Terra is 20% cheaper, and GPT-5.6 Sol gains a faster API path. The July 30 announcement is aimed at businesses that need to decide how much intelligence, speed and reliability each workflow actually requires. The practical takeaway is to route work by outcome and urgency instead of sending every task to the same model tier.

Definition: GPT-5.6 now combines lower Luna and Terra prices with a premium Fast mode for Sol in the API.

Example: A team could reserve Sol for uncertainty-heavy planning, use Terra for routine tool workflows and send high-volume first passes to Luna—if evaluations show the cheaper tier meets the same acceptance bar.

Key takeaway: The announcement changes the economics of model routing more than it changes the model lineup.

Business impact: Lower-cost inference can make more repetitive work economical, while Fast mode gives time-sensitive Sol workloads a paid speed option.

What changed in GPT-5.6 pricing?

GPT-5.6 Luna now costs 80% less and GPT-5.6 Terra costs 20% less than before, according to OpenAI's July 30 update. The reductions apply to API customers and are also reflected in how Luna and Terra usage counts against paid subscriptions in Codex and ChatGPT Work. For a business running high-volume classification, document processing or routine implementation, the immediate action is to recalculate cost per accepted task rather than treating the discount as a reason to switch every request automatically.

GPT-5.6 Luna and GPT-5.6 Terra give operators a wider low-cost range while GPT-5.6 Sol remains the high-capability reference point. The pricing table matters when a workflow repeats thousands of times, because a small per-task difference can determine whether broad automation is economical. The new prices do not prove that Luna or Terra will match Sol on every task, so teams should compare accepted quality, retries and total token use before changing a production route.

What does Fast mode change for GPT-5.6 Sol?

GPT-5.6 Sol Fast mode offers up to 2.5× faster speeds than Standard processing at twice the price, with no change in intelligence. OpenAI says Fast mode replaces Priority Processing in the API, while existing requests tagged priority continue to work and automatically use the new path. The choice is therefore a latency-versus-price decision: use Fast mode when waiting has a measurable business cost, and keep Standard processing for work where the same result can arrive later.

GPT-5.6 Sol Fast mode extends the price-performance decision beyond model selection into response-time selection. A customer-facing agent, incident-response assistant or time-sensitive coding loop may justify the premium when faster completion changes the outcome; a background batch does not automatically do so. The concrete test is to measure whether the faster response improves throughput, conversion, resolution time or another defined outcome enough to cover the additional processing price.

Why does OpenAI say GPT-5.6 is more efficient?

OpenAI attributes GPT-5.6's efficiency to improvements across the model, inference stack and agentic harness rather than to one isolated optimization. The company describes better routing, more efficient token generation and context management that helps agents avoid repeating completed work. For an operator, the useful implication is that cost should be measured across the whole workflow—including model calls, context, tools and retries—because serving efficiency can change the economics of a multi-step agent even when the task itself stays the same.

OpenAI says GPT-5.6 Sol helped reduce serving cost by 20% and increase token-generation efficiency by more than 15% during a human-led engineering process. OpenAI says Sol autonomously rewrote and optimized production kernels, ran hundreds of experiments and monitored training while people remained involved. These figures describe OpenAI's reported internal engineering work, not a universal deployment guarantee; a business should treat them as an explanation for the price changes and verify its own latency, token and infrastructure measurements.

How should teams match GPT-5.6 to a workflow?

GPT-5.6 model choice should follow the required outcome, error cost, urgency and scale of each workflow. OpenAI recommends defining the quality standard first, then using evaluations to determine where extra intelligence materially improves the result and where faster, cheaper processing delivers the same quality. A coding workflow might use Sol to resolve uncertainty and define a plan, then use Luna to implement well-specified changes and run tests; that routing pattern should be adopted only when the team's own evaluation supports it.

GPT-5.6 makes staged routing more practical when a business measures the accepted result at every step. The earlier GPT-5.6 launch introduced Sol, Terra and Luna as distinct capability tiers; this update changes the cost and latency trade-offs between them. Teams should keep the acceptance criteria fixed, compare the current route with Sol, Terra and Luna, and include retries, tool calls, output tokens and review time in the comparison before changing defaults.

What is available now?

GPT-5.6 Luna, Terra and Sol remain available in ChatGPT Work, Codex and the OpenAI API, with access depending on the product and plan. OpenAI says Free and Go users can access Terra in ChatGPT Work and Codex, while Plus, Pro, Business and Enterprise users can choose Terra and Luna; the API exposes the model family under its own pricing and processing controls. Teams should verify the options in the account that runs their workflow instead of assuming that a subscription label guarantees every model or mode.

The GPT-5.6 update changes model economics without changing ChatGPT or Codex subscription prices and quota budgets. OpenAI says Luna and Terra now consume fewer credits, and the pricing changes will begin rolling out in AWS later on July 30. The operational takeaway is to recheck both API invoices and product-level usage accounting after rollout, because lower list prices and lower subscription credit consumption affect different parts of the same AI budget.

What should operators watch next?

The important GPT-5.6 evidence after this announcement will be workflow-level cost and quality, not the size of the headline discount. OpenAI's update expands the range from low-cost Luna work to premium-speed Sol work, but the right balance depends on the accepted outcome a business needs. Teams should test one or two repeatable workflows, record quality, latency, retries and total spend, and expand only when the measured result shows that the new tier or processing mode creates a durable operational gain.