OpenAI has launched GPT-6 Sol and GPT-6 Luna, extending the GPT-6 family with lower-cost models aimed at everyday professional work, coding and agentic tasks. The company has set Sol API pricing at $2 per million input tokens and $10 per million output tokens, while Luna is priced at $0.10 and $0.50 respectively. OpenAI says both represent 50 per cent reductions against GPT-5.6 promotional pricing.
The launch is heavy on performance and cost comparisons. OpenAI reports gains on professional workflow, coding, factuality and computer-use evaluations, and says improved prompt caching can discount cached input reads by 90 per cent. It also cites customer and benchmark results comparing its models with Anthropic systems. Those figures are vendor-reported and should not be treated as neutral procurement evidence without workload-specific testing.
The enterprise story is less glamorous than the benchmark table. Cheaper inference and better cache reuse make it economically possible for agents to run for longer, retain more context and attempt more steps. That raises the value of evaluation because a small failure rate can become more consequential when multiplied across longer workflows. Cost, reliability and control are therefore converging into the same procurement decision. A cheaper agent is useful only if the organisation can still see what it is doing and stop it when required.
