GPT-6 Sol and Luna Target Cheaper AI Coding and Enterprise Agents

SAN FRANCISCO — OpenAI has restructured its mid-tier model portfolio with GPT-6 Sol and GPT-6 Luna, two high-efficiency systems designed to reduce the cost of automated programming and enterprise agent workflows.

The models were made available on day one through Codex and ChatGPT Work for accounts under Plus, Pro, Edu, Business, and Enterprise subscription tiers. Access across the standard ChatGPT conversational web interface is scheduled to activate incrementally.

GPT-6 Sol is priced at $2.00 per million input tokens and $10.00 per million output tokens. GPT-5.6 Sol previously cost $4.00 per million input tokens and $20.00 per million output tokens, while access through OpenAI’s Application Programming Interface (API) reflects a 50 percent price reduction relative to the promotional rates previously offered for GPT-5.6 Sol and Luna models.

For high-volume, low-complexity applications, GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens. Those rates compare with $0.20 per million input tokens and $1.20 per million output tokens for GPT-5.6 Luna.

The new variants are lower-cost alternatives to OpenAI’s flagship GPT-6 Astra model. Optimization techniques developed during the construction of Astra are incorporated into Sol and Luna, which are intended to provide developer-grade performance for routine software engineering and continuous backend tasks while substantially reducing per-token computing overhead.

OpenAI reported that architectural refinements allow the models to deliver capabilities comparable to GPT-5.6 Sol while running at approximately 1 percent of its original execution cost. Input tokens encompass the instructions and context supplied to a system, while output tokens represent the text or code generated in response. Output pricing is therefore critical for code generation and automated debugging.

Internal evaluations show that GPT-6 Sol and Luna cut output error rates by approximately 50 percent compared to the GPT-5.6 baseline. Their development focused in part on retaining accuracy during long-context processing, and decreased hallucination rates are particularly critical in software development, where minor code errors can disrupt operational pipelines.

Performance metrics released by OpenAI indicate that GPT-6 Sol outperforms competing systems, including Anthropic’s Claude Fable 5.1, while operating at a lower price point. Benchmark testing also indicates that GPT-6 Sol yields higher accuracy on select evaluation suites than low-effort reasoning configurations of GPT-6 Astra.

The release comes as businesses move from simple chat interfaces toward autonomous agentic deployments, in which systems run uninterrupted loops to execute multi-step tasks. Token processing fees have emerged as a primary bottleneck for enterprise adoption, alongside a broader shift in enterprise AI strategy toward inference economics as well as pure model capability.

To prevent service bottlenecks similar to those experienced during the initial deployment of GPT-6 Astra, OpenAI executed a multi-channel rollout across its enterprise and developer platforms. Non-paying users and subscribers on the ChatGPT Go tier are restricted to GPT-6 Luna through the platform’s desktop application.