New model releases arrive weekly, and price per task is the real fight

Claude Haiku 5.5 landed (49996437) with tiered pricing that people called 'a bit weird': $0.10 per million input tokens under 100K, $0.50 above. A Sonnet cache-read price cut got a cheer. OpenAI announced 'GPT-6 and Intelligent UI for everyone' (49996425). Someone gave Opus 5.5 one prompt and six hours to visualize Invisible Cities (50004790), and the top comment ignored the art and compared cost: $74 vs $10 vs $25 for the same prompt.

The through-line is that people have stopped asking whether the model can do it and started asking what it costs. 'What can we get per dollar' with six subagents running in parallel is now the framing. One commenter asked if there's an LLM equivalent of Moore's Law, which says how fast the ground is moving.

Pricing that depends on context length means your architecture decisions, such as how much you stuff into a prompt, directly set your margins.