OpenAI launches GPT-6 Sol and Luna at lower cost
OpenAI has launched GPT-6 Sol and GPT-6 Luna, bringing GPT-6 Astra capabilities to faster, more affordable models for work at scale.
OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22, positioning the pair as faster, lower-cost members of the GPT-6 family. The models carry forward capabilities introduced with GPT-6 Astra while targeting work at scale, where latency and inference cost matter as much as peak performance.
What GPT-6 Sol and Luna change
OpenAI describes Sol and Luna as two different trade-offs within the same generation. Sol is aimed at higher capability, while Luna is designed for lower cost and faster throughput. The company says both models inherit advances in professional work, factuality, coding, computer use, and alignment from Astra.
That makes this a model-portfolio change rather than a single flagship refresh. Teams can choose a cheaper model for routine volume and reserve the more capable option for tasks that justify extra latency or spend.
Access across ChatGPT, Codex, and the API
OpenAI’s launch material says GPT-6 Sol and Luna are available to ChatGPT Plus, Pro, Business, Enterprise, and Edu customers in ChatGPT Work and Codex. Free and Go users receive access to GPT-6 Luna. OpenAI Developers also announced API availability and said the launch prices are 50% lower than GPT-5.6.
The exact model IDs, rate limits, and rollout conditions still belong to the relevant product documentation. Developers should check the API model catalog and their account’s availability before changing production routing, especially where a staged rollout can expose different limits by plan or region.
Why the pricing change matters for agents
Lower unit cost changes the economics of agent workflows. Classification, retrieval, planning, code review, and background research can run more often when the model price leaves room for retries and longer context. That does not remove the need for evaluation: cheaper calls can still create higher total spend if a workflow loops or expands its context unnecessarily.
OpenAI also published a separate update on GPT-6 prompt caching, describing higher cache-hit rates, new diagnostics, explicit breakpoints, and controls intended to reduce latency and cost. Together, the model and caching announcements point toward a broader optimization for repeated, tool-using workloads rather than a benchmark-only upgrade.
What to verify before switching
Compare Sol and Luna on representative tasks, not only headline benchmarks. Measure factuality, tool-call reliability, latency, cached-input behavior, context limits, and the total cost of a complete workflow. Keep a fallback route during rollout, and confirm whether current API aliases or ChatGPT workspace settings select the new models automatically.
OpenAI’s RSS announcement is the primary record of the launch. TechCrunch independently lists the release as a lower-cost, lower-error GPT-6 update. The practical next milestone is documentation-level clarity on stable API identifiers, regional availability, and the limits attached to each plan.
Sources
Try the related loot
Give Any Model a Sandboxed Shell and File Workspace with OpenRouter