Intelligence now has a price list, and the floor just dropped to $0.10
In one month, frontier AI split into clear price tiers. Choosing the right tier is now a cost lever, not a technical detail.
September brought a wave of model launches from every major lab. The most important shift was in pricing. On September 22 OpenAI released two smaller GPT-6 models, one of them priced at $0.10 per million input tokens, and Anthropic launched a cheaper new flagship the same day.
What happened
OpenAI now sells three tiers. Luna, at $0.10 input and $0.50 output per million tokens, is for high-volume, narrow tasks. Sol, at $2 and $10, is for everyday knowledge work. Astra, at $10 and $50, is for the hardest problems. The two smaller models cost roughly half as much as the versions they replace. Across the market, the gap between the cheapest and most expensive leading models is now about 100x.
Why it matters
Most companies still send every task to one ‘best’ model. That's like having partners do data entry. Routine work such as classification, extraction and summarisation now costs next to nothing. Top-tier reasoning is still expensive.
What it means for your business
Map your AI workloads by difficulty and volume, then route each one to the cheapest model that clears the quality bar. In high-volume operations like customer service, back office and document processing, this is often the fastest route to a better AI business case.
Sources: VentureBeat, The Next Web, Artificial Analysis (Sep 22); KDnuggets (Sep 22)