DeepSeek released V4 Flash on Friday, a model that performs near Claude Opus 4.8 on coding benchmarks at roughly 1 percent of the cost [Axios]. OpenAI reduced GPT-5.6 Luna pricing by 80 percent the prior day after launching the model three weeks earlier [Axios]. Analyses differ on whether the price compression primarily expands access, enforces market discipline, or raises verification and security questions.