OpenAI cut the price of GPT-5.6 Luna, its self-described fastest and most affordable model, by 80 percent, and Anthropic launched Claude Opus 5 at half the price of its flagship Fable 5. Together the moves have pulled token prices from leading US labs down almost a quarter since mid-July, according to Silicon Data's token price index. Both companies are plotting IPOs at trillion-dollar valuations while investors hunt for evidence that the industry's spending can throw off returns.
OpenAI dropped GPT-5.6 Luna from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens. Anthropic priced Opus 5 at $5 per million input tokens and $25 per million output tokens, positioned explicitly at half the cost of Fable 5. This week Anthropic also called off a planned September price increase on Sonnet 5.
The immediate pressure is coming from China. Moonshot's Kimi K3 and DeepSeek's V4 Flash have closed the performance gap enough that cost-conscious buyers are willing to switch, and they are switching from inside Silicon Valley itself. DoorDash and Airbnb have both said they have started using Chinese-made models to rein in AI bills.
Key facts
- 01OpenAI cut GPT-5.6 Luna input tokens from $1 to $0.20 per million and output from $6 to $1.20 per million — an 80% reduction.
- 02Anthropic launched Claude Opus 5 at $5 per million input tokens and $25 per million output, half the price of its Fable 5 flagship.
- 03Token prices from leading US labs have fallen almost 25% since mid-July, according to Silicon Data's index.
- 04DoorDash and Airbnb have started using Chinese-made models to rein in AI bills.
- 05Both OpenAI and Anthropic are plotting IPOs at trillion-dollar valuations.
That switch is happening at a moment when Anthropic and OpenAI are shifting enterprise customers away from flat subscriptions toward usage-based billing, where the meter runs on tokens consumed. Some corporate users have responded by capping AI usage internally, others by piloting cheaper alternatives outright. When enterprise finance teams see the invoice climb, the open Chinese model that a developer can download, tweak, and self-host starts looking a lot more attractive than a proprietary API.
Artificial Analysis, which benchmarks models on math, science, coding, and reasoning, found Opus 5 at medium effort delivered similar performance and cost per task to Kimi K3 at max effort. GPT-5.6 Luna at max effort performed similarly to DeepSeek V4 Flash at max, but cost just under twice as much per task. In other words, US mid-tier models are now roughly at parity on quality with Chinese open competitors, but the Americans are still charging a premium — which is exactly what the recent cuts are designed to close.
“The US labs have cut the middle and are defending the top.”— Mantas Lukauskas, AI tech lead at Hostinger
Mantas Lukauskas, AI tech lead at website host Hostinger, which has run large language models in production since 2020, said prices for the very best models were flat to rising even as the middle collapses. He framed the current moment as the first real test of whether Anthropic and OpenAI can protect the price of their most advanced offerings while ceding ground below.
Anthropic and OpenAI declined to comment on the competitive framing. A person close to Anthropic pushed back on the read that Opus 5's pricing was a response to Chinese rivals, saying the below-flagship price point simply reflected how the
“family of models is built, so there's no connection to competitors.”— Person close to Anthropic, Anthropic source
The nuance the price sheet hides is that headline token cost isn't the same as cost per task. More capable models can finish a job in fewer tokens or fewer attempts, and most modern models expose an effort setting that trades compute for accuracy. A model that looks expensive per token can be cheaper per completed job — which is the argument the US labs will lean on hardest as the pricing conversation moves to boardrooms.
Still, the trajectory is unmistakable. A nearly 25 percent drop in effective prices from the frontier US labs in about a month is the sharpest downward move the market has seen, and it happened because customers had a real alternative. Open Chinese weights change the negotiation. When DoorDash and Airbnb are on the record about switching, every other enterprise procurement team gets to point at that filing in its next contract discussion.
The strategic question for OpenAI and Anthropic on the road to trillion-dollar listings is whether the top of the stack — the truly frontier models where American labs still lead — throws off enough margin to fund the capex, or whether the middle tier they just torched was carrying more of the P&L than they'd like to admit. Cutting the middle and defending the top is a coherent strategy only if the top holds. Kimi K3 and DeepSeek V4 Flash are already close enough at mid-tier that the frontier gap is the only pricing power left, and Chinese labs have not shown any inclination to stop closing it.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
