Anthropic's September 2026 launch of Claude Opus 5.5 drove recent sentiment by cutting listed output token pricing to $20 per million tokens—20% below the prior Opus 5 rate—while delivering 30%+ faster generation and cache reads at just $0.20 per million. Efficiency gains, including adaptive thinking and prompt caching, reportedly reduce typical workload costs by 40% even as output volume rises with agentic use. Competitive pressure from Google's Gemini 4 Argon, set to match the $20 rate after its introductory period, and OpenAI's tiered GPT-6 offerings reinforces downward momentum on frontier large language model pricing. Traders are watching whether further model iterations or inference optimizations push rates lower before year-end or stabilize near current levels amid sustained demand for high-capability reasoning.