The frontier just got cheaper

Same Tuesday: Anthropic ships Opus 5.5 for less; OpenAI answers with GPT-6 Sol and Luna at half the old token price. Xiaomi’s open weights and Jev already punched the mid-stack. The race did not pause — it cut the bill.

By Drew Wall, September 22, 2026

Two weeks after lab leaders asked to pace the frontier, the product calendar did the opposite. On September 22, Anthropic shipped Claude Opus 5.5 with a price cut — and about ninety minutes later OpenAI answered with GPT-6 Sol and Luna at roughly half the GPT-5.6 token rates. The week's heat was not a new AGI claim. It was a cheaper bill.

Same day, two price cuts

Opus 5.5 lands at $4 / $20 per million input/output tokens (down from $5 / $25), with Anthropic pitching agentic coding, clearer writing, and fewer boundary-dodging attempts. GPT-6 Sol is $2 / $10; Luna is $0.10 / $0.50 — permanent list prices, not a weekend promo. Sol is the economical agent / coding tier under Astra; Luna is the high-volume workhorse for extract, summarize, classify. Artificial Analysis says cost-per-task roughly halves versus the 5.6 twins while intelligence scores stay in the same band. Head-to-head Sol vs Opus 5.5 is still thin; the marketing war is already thick.

Pressure from below

The closed labs are not the only ones moving the floor. Xiaomi's MiMo-V2.6-Pro put open weights near the top of the Artificial Analysis Intelligence Index at a fraction of Grok-class token prices — MIT license, self-hostable. TypeSafe's Jev keeps arguing that many production decisions do not need a frontier chat model at all. Independent checks have cooled some of Jev's wildest multiples, but the directional point stands: if mid-stack work gets good enough and cheap enough, Astra- and Fable-tier premiums only hold for the jobs that truly need them. Related: Open-weight models; AI profitability crisis.

The point

Prices matter more than the talk about slowing down. If GPT-6 Sol and Claude Opus 5.5 make it cheap enough to run agents all day, buyers will stop picking one flagship model and start matching model tiers to tasks. The big labs are cutting list prices, while open-weight and specialized models keep pushing the lowest price down. Related: Language Models; Agent Frameworks; The AI slowdown hit the capex trade.