OpenAI introduced GPT-6 Sol and GPT-6 Luna, cutting their API prices by up to 50% from GPT-5.6 promotional rates, hours after Anthropic launched Claude Opus 5.5 and said it costs 40% less to run than its predecessor. The back-to-back releases highlight an industry shift toward cost-per-task and efficiency as companies deploy AI agents for coding, research, computer use and other extended workflows. Sol costs $2 per million input tokens and $10 per million output tokens, while Luna costs $0.10 and $0.50, respectively. OpenAI positioned GPT-6 Astra as its highest-performance model and said Sol scored 33.2% on AutomationBench at a reported cost of $0.27 per task, compared with Astra's 30.3% score at low effort and 3.9 times the cost. Anthropic said cache reads represent most costs in agentic and coding workloads and cut its cache-read price by 60% from Opus 5. Earlier information in the topic record said Opus 5.5 was released at 16:31 UTC, followed by OpenAI's models at 18:12 UTC, 101 minutes later, and that both launches came 10 days after Dario Amodei advocated slower capability gains, a position endorsed by Sam Altman and supported by Elon Musk.