OpenAI Slashed GPT-5.6 Luna's Price 80% as Chinese Models Hit 46% of US Enterprise Tokens, and Investors Moved $3 Billion Into Vertical Agents Instead of Chat

Three numbers from the last two weeks tell the same story from different angles: 80%, 46.4%, and $3.07 billion. OpenAI cut its cheapest model's price by the first, Chinese open-weight models now carry the second share of US enterprise AI traffic, and investors just funneled the third into agents that do one job well instead of chatbots that try to do everything. None of these is a coincidence. Here's what happened, and what it means for anyone deciding what to build on top of a model this quarter.

1. GPT-5.6 Luna Gets an 80% Price Cut, Three Weeks After Launch

On July 30, OpenAI dropped GPT-5.6 Luna's price from $1/$6 to $0.20/$1.20 per million input/output tokens — an 80% cut on the model it had shipped just three weeks earlier, on July 9. Terra, the mid-tier model, got a smaller 20% cut, from $2.50/$15 to $2/$12. The flagship Sol model held steady at $5/$30. OpenAI framed the move as passing along efficiency gains — better speculative decoding, tighter context management, a training pipeline that increasingly uses GPT-5.6 itself to optimize serving code. That explanation is plausible, but the timing lines up too neatly with a second story to be the whole picture.

Cutting the cheap, high-volume tier hardest while leaving the flagship untouched is a defensive move, not an efficiency dividend — it's the tier most exposed to substitution that got repriced.

2. Chinese Models Now Move Nearly Half of US Enterprise Tokens

A CNBC investigation published in early July found that Chinese-origin models have held at least 30% of OpenRouter's weekly enterprise token volume since February, peaking at 46.4% — versus 35.7% for US-origin models — up from just 4.5% a year earlier. DeepSeek alone accounts for 17.6% of routed tokens (roughly 5.13 trillion a week), the single largest vendor on the platform; Alibaba's Qwen follows at 13.9% (2.77 trillion). The draw is price: DeepSeek V4-Flash runs about $0.14 per million input tokens against GPT-5.5's $5.00, a gap open-weight Chinese labs have sustained by pricing 60–90% below US frontier labs since February. Luna's new $0.20 input price closes that gap dramatically — it's now roughly in the same neighborhood as DeepSeek rather than 30x above it.

3. The Money Is Skipping Chat and Going to Vertical Agents

While chat pricing collapses, venture money is moving somewhere chat commoditization can't touch it: agents scoped to a single regulated workflow. Vertical AI deals pulled in $3.07 billion across 73 rounds over the trailing year, with legal, insurance, construction, and healthcare together taking roughly three-quarters of that capital on about half the deal count. Legal AI leads by dollars — around $1 billion so far in 2026, a third of total vertical AI capital — while finance AI is the sharpest breakout, jumping from a single comparable deal in 2025 to 16 deals and about $515 million in 2026. Healthcare's capital share fell from 56.7% to 27.1% even as its deal count stayed the highest of any category, a sign the category is maturing from a few megarounds into many smaller, provable deployments rather than cooling off.

What Connects the Three

Read together, these aren't separate stories about pricing, geopolitics, and venture capital — they're the same market repricing the same thing from two directions at once. General-purpose chat is racing toward zero marginal cost because DeepSeek and Qwen made "just answer questions well" a commodity that OpenAI now has to price-match rather than out-compete. At the same moment, the capital that would once have funded another general chatbot is instead rewarding teams that picked one workflow — legal discovery, claims processing, financial reconciliation — and built something a regulator or a compliance team can actually sign off on. If your roadmap this quarter is "a better chat interface," the ground under that plan just moved twice in two weeks. If it's "an agent that owns one workflow end to end," both of these trends are blowing in your favor.

— Maya