OpenAI unveils GPT-6 Sol and Luna with API prices cut in half

On 22 September 2026, OpenAI introduced two new variants of its sixth‑generation language model: GPT‑6 Sol, aimed at complex tasks such as code generation, and GPT‑6 Luna, optimized for large‑scale, well‑defined operations like summarisation and information extraction. The launch follows GPT‑5.6 by less than three months and arrives a few weeks after GPT‑6 Astra.

Pricing strategy halves previous costs

OpenAI disclosed that the API rates for both GPT‑6 Sol and Luna are set at exactly 50 % of the prices charged for GPT‑5.6 Sol and Luna. The company attributes the reduction to advances in caching mechanisms and more efficient inference pipelines, which together lower the computational overhead per request.

The new pricing model applies across all OpenAI delivery channels, including ChatGPT Work, the Codex suite, and the public API. Luna is also slated for inclusion in the desktop application and the Free and Go subscription tiers, widening its accessibility for users with modest budgets.

Error‑rate improvements claimed by OpenAI

According to a TechCrunch report, OpenAI claims that GPT‑6 Sol generates roughly half the factual errors of its predecessor when evaluated on an internal benchmark that aggregates user‑reported conversation mistakes. Golem.de echoes this, stating that the error rate for Sol dropped from about 20 % to 10 % in OpenAI’s internal testing.

For Luna, OpenAI reports that its error performance matches that of GPT‑5.6 Sol while costing only a fraction of the price. This suggests that Luna delivers the same reliability as the older, more expensive model, but with the added benefit of the new pricing structure.

• Half‑price API for both models
• Reduced factual error rate for Sol
• Luna’s error rate comparable to GPT‑5.6 Sol
• Availability in Free and Go plans

Market context and competitive timing

The announcement came roughly 90 minutes after Anthropic released Claude Opus 5.5, highlighting a rapid escalation in the performance‑to‑price race among leading AI providers. Both firms are positioning their latest models as the most cost‑effective options for enterprise and developer workloads.

OpenAI emphasizes that the reported error reductions are based on internal measurements and have not undergone independent, exhaustive validation. The company cautions that real‑world performance may vary depending on the specific use case and data quality.

Despite the lack of third‑party verification, the announced improvements could influence procurement decisions for companies that balance model accuracy against operational spend. The dual‑track approach—Sol for complex, code‑heavy tasks and Luna for bulk processing—offers a clearer cost‑benefit matrix.

The rollout also expands the ecosystem of OpenAI products, with Luna slated for integration into the upcoming desktop client, potentially simplifying workflow automation for users who prefer on‑premise tools.

For English‑speaking organisations, the concrete impact is twofold: first, the halved API fees translate into immediate budget relief for teams that already rely on OpenAI’s models for development, content creation, or data extraction. Second, the lower error rates—particularly for Sol—reduce the need for extensive post‑processing or human review, allowing faster iteration cycles and higher confidence in automated outputs.

Implications for developers and enterprises

Developers working on code‑intensive applications can now tap into GPT‑6 Sol’s enhanced accuracy without the previous cost barrier, making it feasible to embed sophisticated code suggestions directly into IDEs or CI pipelines.

Enterprises focused on large‑scale document handling stand to benefit from Luna’s unchanged reliability paired with a 50 % price cut, which could shift budgeting decisions away from legacy on‑premise NLP stacks.

Future outlook

Analysts anticipate that OpenAI’s aggressive pricing will pressure competitors to accelerate their own cost‑reduction strategies, potentially sparking a new wave of innovation in model optimisation and hardware efficiency.

The company has hinted at further refinements to both Sol and Luna later in the year, suggesting that the current launch is part of a broader roadmap aimed at consolidating OpenAI’s market leadership.

In Japan, where the announcement was published in both English and Japanese, local startups and tech giants alike are already evaluating the new models for integration into customer‑service chatbots and automated reporting tools, reflecting the regional appetite for affordable, high‑quality AI services.