OpenAI Deploys GPT-6 Sol and Luna at Half Anthropic Opus 5.5 Price Hours After September 22 Release

Anthropic unveiled Claude Opus 5.5 on September 22 at $4 input and $20 output per million tokens, claiming 40 percent cost reduction from Opus 5 and leading scores on Terminal-Bench 4.0 at 66.4 percent. OpenAI responded the same afternoon with GPT-6 Sol priced at $2 input and $10 output, plus 90 percent cache discounts, and Luna at $0.10 input and $0.50 output. Both firms referenced external evaluations from METR and Frontier Design, yet neither published full task-level token consumption data that would allow direct cost-per-completion comparison. The sequence followed Dario Amodei's September 12 call for pacing frontier development. Primary records show both labs instead accelerated releases, with OpenAI's cached pricing directly targeting agentic workloads that Anthropic identified as its core use case. Usage-limit increases of 20 percent on Anthropic Pro plans and OpenAI's footnote on safeguard fallback rates indicate competitive positioning rather than coordinated restraint. OpenAI's charts omitted Opus 5.5 results while Anthropic's statement highlighted 30 percent faster output generation. The documented price trajectory from $4/$20 to $2/$10 within a single trading day demonstrates that stated safety coordination yields to immediate market-share defense when token volume is at stake. Next moves center on whether cached-read discounts expand to non-agentic tasks and whether usage caps on both platforms remain stable through October. Sustained price compression below $2 input would require either larger training runs or reduced safety evaluation overhead, both of which carry measurable compute and regulatory costs.