OpenAI slashes AI compute costs with GPT-6 Sol and Luna models

OpenAI has officially expanded its flagship model lineage with two new releases: GPT-6 Sol and GPT-6 Luna. Designed to substantially lower the barrier to entry for enterprise users, these releases deliver top-tier performance while cutting API processing costs by half compared to the previous GPT-5.6 benchmarks. By introducing these targeted model tiers underneath the core capability threshold of GPT-6 Astra, OpenAI is enabling organizations to tailor their computational overhead and power consumption directly to the specific technical demands of each application.

The dual launch reflects a strategic segmentation between high-level logical reasoning and high-volume background execution. GPT-6 Sol takes center stage as the heavyweight engine built for complex problem-solving and demanding corporate workflows. Priced at 2 dollars per million input tokens, it maintains sophisticated reasoning capabilities at an operational fraction of legacy prices. Meanwhile, GPT-6 Luna targets high-throughput routine processing at a razor-thin 0.10 dollars per million input tokens. Built specifically for rapid execution across massive user queries, Luna serves as an ideal backend for automated customer support frameworks, real-time context translation, and lightweight system automation.

Industry benchmarks confirm that the dramatic drop in pricing does not compromise performance metrics. On AutomationBench—a test designed to evaluate AI agent efficiency across multi-step software tool chains—GPT-6 Sol set to maximum reasoning effort achieved a impressive 33.2 percent success rate. This score edges past competing high-tier solutions like Anthropic’s Claude Opus 5, while operating at roughly 9 percent of the rival’s total per-task expense.Concurrently, GPT-6 Luna recorded a 5.4 percentage point gain over its predecessor in automation metrics while reducing individual action costs by 58 percent.

Code synthesis and repository analysis formed a cornerstone of this architectural upgrade, addressing the reliance of contemporary developers on autonomous engineering assistants. On the rigorous DeepSWE v1.1 evaluation framework—which measures real-world problem resolution across large-scale software engineering bases—GPT-6 Sol posted a 68.8 percent resolution rate under peak effort parameters. This performance directly rivals top-tier competitors like Claude Fable 5 while dramatically undercutting market costs. Simultaneously, the lightweight GPT-6 Luna successfully rivaled mid-tier configurations of Anthropic models, completing test pipelines at a 93 percent cost reduction compared to Opus 5.

By systematically lowering operational overhead without forfeiting reasoning depth, OpenAI’s latest releases signal a broader trend toward sustainable, cost-efficient deployment of foundational models across corporate environments.