OpenAI and Anthropic Model Competition Shifts Toward Cost per Task

Sam Altman claims GPT-6 Sol and Luna improve intelligence, alignment, coding, computer use, and work output at lower cost; Dan Shipper's firsthand Codex testing offers an anecdotal comparison with Astra and Opus 5.5. Podcasts and recent commentary reinforce a shift toward coding throughput, agent economics, and enterprise deployment, but pricing and capability claims remain thinly documented.