OpenAI Halves the Cost of GPT-6 Sol and Luna for Businesses

30-Second Summary

OpenAI launched GPT-6 Sol and Luna on 23 September 2026, completing the initial rollout of its GPT-6 model family. Both models arrived with API prices cut by 50 per cent or more versus the previous GPT-5.6 generation. GPT-6 Sol, the mid-tier model designed for everyday professional and coding tasks, now costs $2 per million input tokens and $10 per million output tokens. OpenAI says Sol makes approximately half as many factual errors as its predecessor, bringing it close to GPT-6 Astra quality at significantly lower cost. All paid ChatGPT tiers and API customers gained access on the day of launch, with Free and Go users receiving access to Luna on the desktop app.

At a Glance

• Topic: Model Releases
• Company: OpenAI
• Date: 23 September 2026
• Announcement: GPT-6 Sol and Luna launched with 50 per cent lower API pricing than the GPT-5.6 generation
• What Changed: Both models are faster, more accurate, and cost half as much per API call
• Why It Matters: The cost calculation for AI-assisted business workflows has fundamentally shifted
• Who Should Care: Any business using ChatGPT Work, the OpenAI API, or evaluating AI tools for operational workflows

Key Facts

• Company: OpenAI
• Launch Date: 23 September 2026
• What Changed: GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for GPT-5.6 Sol. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, down from $0.20 and $1.20
• Who It Affects: ChatGPT Plus, Pro, Business, Enterprise, and Edu subscribers; OpenAI API customers; Free and Go users (Luna on desktop)
• Primary Source: OpenAI blog and VentureBeat (23 September 2026)

What Happened

OpenAI released GPT-6 Sol and Luna on 23 September 2026, filling out the GPT-6 model family below the previously launched GPT-6 Astra flagship. Both models launched with permanent API price reductions of at least 50 per cent compared to their GPT-5.6 predecessors. OpenAI attributed the cuts to structural improvements in caching and inference efficiency rather than a promotional campaign.

GPT-6 Sol is positioned as the mid-tier everyday model, with a 1,050,000-token context window and support for text and image input. OpenAI says Sol makes approximately half as many factual errors as GPT-5.6 Sol, with independent testing showing maximum-effort hallucination rates dropping from 92 per cent to 60 per cent. The company describes Sol as approaching GPT-6 Astra capability on general tasks while costing significantly less to run.

GPT-6 Luna occupies the lightweight, high-volume tier, priced at $0.10 per million input tokens. Luna is now accessible to ChatGPT Free and Go users through the desktop application, as well as to all paid tiers through the API. Both models are rolling out gradually to ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu subscribers, with API access available immediately.

OpenAI said the pricing reductions reflect efficiency gains from improved caching and inference optimisation, suggesting further reductions may follow as the company refines its infrastructure.

Why It Matters

• The 50 per cent price reduction on Sol moves the cost-per-task calculation for AI-assisted workflows into territory where it is cheaper than the time saved in many common business processes.
• GPT-6 Sol's improved accuracy, roughly half the factual error rate of its predecessor, makes it more suitable for customer-facing content, research outputs, and any task where an incorrect answer creates a problem.
• The tiered model structure (Astra, Sol, Luna) gives businesses a clear routing strategy: high-stakes complex tasks to Astra, everyday professional work to Sol, and high-volume lightweight tasks to Luna.
• OpenAI attributed the pricing change to structural efficiency gains, suggesting further reductions are possible as infrastructure matures.
• Luna access for Free and Go ChatGPT users expands the number of employees who can experiment with AI assistance, lowering internal adoption barriers for teams that have not yet embedded AI into daily work.
• The move continues a pattern across the industry where frontier AI has become substantially cheaper over 12 months, driven by competition from open-weight models and Chinese AI providers.

The David and Goliath View

OpenAI's price cuts matter most to the businesses that were on the margin of making AI work economically. For a company that processes hundreds of documents a month, routes customer enquiries through an AI assistant, or uses AI to draft and personalise sales outreach, the cost model at $2 per million tokens for Sol is now very different from what it was at $4. For a company using Luna at $0.10, the economics of high-volume, lightweight AI tasks have crossed into near-zero marginal cost territory.

The risk for lean operators is not that they pay too much. It is that they wait. The businesses that built workflows on GPT-5.6 at higher prices are already ahead. At lower prices, those same workflows become available to smaller teams with tighter margins. This is the moment to revisit any AI use case that was shelved because the numbers did not quite work, to rerun the cost calculation with the new pricing, and to move from evaluation to implementation on at least one operational workflow.

The immediate recommendation is practical: if you have a ChatGPT Work or API subscription, calculate what you spent on AI calls last month and confirm your projected spend at the reduced rates. Then identify one workflow where you were limiting AI usage because of cost, and remove that constraint.

Where This Fits in the AI Stack

AI Growth Engine: Lower costs make it economically viable to run AI across more touchpoints in the sales and marketing process, from research and personalisation to content generation at scale.

Employee Amplification Systems: At half the previous API cost, AI assistance can be embedded into more employee workflows without the per-seat or per-call expense becoming a budget conversation.

Questions Operators Are Asking

Do I need to do anything to get the lower pricing? For OpenAI API customers, the new pricing applies automatically to calls made with the GPT-6 Sol or Luna model IDs from 23 September 2026. If your codebase still references GPT-5.6 model IDs, you will need to update those to access the new models and lower prices. For ChatGPT Work subscribers, the update is rolling out gradually to the ChatGPT interface with no action required.

Is GPT-6 Sol actually better than what I was using, or just cheaper? Both, according to OpenAI's published results and independent benchmarks. GPT-6 Sol makes roughly half as many factual errors as GPT-5.6 Sol, and independent testing confirms a meaningful reduction in hallucination rates. If you are running tasks where accuracy matters, Sol is not just cheaper but more reliable than its predecessor.

When should I use Luna versus Sol? Use Luna for high-volume, lower-stakes tasks where speed and cost matter most, such as tagging, classification, short summaries, or drafts you will review before sending. Use Sol when the output goes to a customer or requires factual precision, such as product descriptions, research summaries, or policy documents.

How does this affect my ChatGPT subscription rather than the API? If you subscribe to ChatGPT Plus, Business, or Enterprise, GPT-6 Sol is being rolled out to the ChatGPT interface for your plan tier. You do not pay extra for the upgrade. The API price reduction is separate from subscription pricing, which has not changed.

Does this change the business case for building custom AI tools? Yes, meaningfully. If you evaluated building an AI-powered internal tool and found the per-call economics marginal, the calculation has changed by roughly half. That is often enough to move a borderline business case from unviable to viable, particularly for tools that process documents, handle customer queries, or assist staff with research and drafting.

Citable Summary

What happened: OpenAI launched GPT-6 Sol and Luna on 23 September 2026, cutting API prices by 50 per cent versus the GPT-5.6 generation while improving factual accuracy by approximately half.

Why it matters: The cost of running AI-assisted business workflows has dropped materially, making AI economically viable for a wider range of operational tasks in lean organisations.

David and Goliath view: The businesses on the margin of making AI work economically now have a new calculation to run, and the moment to move from evaluation to implementation is now.

Offer relevance:

• AI Growth Engine: Lower AI costs enable more affordable deployment of AI-powered sales, marketing, and customer engagement workflows.
• Employee Amplification Systems: Reduced per-call costs allow AI to be embedded across more employee workflows without cost becoming a budget constraint.