OpenAI cut prices on its GPT-5.6 Luna model by 80 percent on July 30, 2026, the sharpest move yet in a pricing war that has now pulled in DeepSeek and reshaped a Google product plan within the same week.
OpenAI drops GPT-5.6 Luna pricing
OpenAI lowered GPT-5.6 Luna from $1 per million input tokens and $6 per million output tokens to $0.20 and $1.20 respectively (VentureBeat, 2026; CNBC, 2026). The company framed the cut as a response to a more cost-sensitive enterprise buyer base and to competition from lower-priced Chinese model providers. OpenAI did not change pricing on its higher-tier GPT-5.6 Sol model in this announcement.
The cut matters because Luna is the model OpenAI positions for high-volume, cost-conscious workloads such as customer support automation and bulk document processing. An 80 percent reduction changes the unit economics for any company running those workloads at scale, not just the sticker price on a rate card.
DeepSeek ships a cheaper model that beats its own flagship
One day later, on July 31, DeepSeek released V4-Flash-0731, a mixture-of-experts model with 284 billion total parameters and 13 billion active parameters, a 1 million token context window, and MIT-licensed weights (MarkTechPost, 2026; Hugging Face, 2026). It is priced at $0.14 per million input tokens and $0.28 per million output tokens.
What makes the release notable is not just the price. DeepSeek reported that V4-Flash-0731 outscored its own larger V4-Pro-Preview model on every published agentic benchmark. On Terminal-Bench 2.1 it scored 82.7 against V4-Pro-Preview's 72.1. On DeepSWE, the smaller model jumped from 7.3 to 54.4, a gain DeepSeek attributes entirely to re-post-training rather than a new architecture (datanorth.ai, 2026).
That combination, a smaller and cheaper model outperforming the company's own premium tier, undercuts the assumption that model size and price track capability in a straight line. It also puts direct pressure on OpenAI's Luna pricing from below.
Google cancels an app one day before its planned launch
On July 31, Google confirmed it was scrapping the standalone AI Studio mobile app for iOS and Android, one day before the app's planned August 1 release (TechRepublic, 2026; Android Authority, 2026). The app had drawn roughly 800,000 pre-registrations across more than 168 countries following its preview at Google I/O in May.
Instead of shipping the standalone app, Google said it will fold AI Studio's capabilities, including app remixing, cross-device project syncing, and a build-test-publish workflow, directly into the Gemini app on both mobile and desktop. Google gave no timeline for when those features will appear inside Gemini, saying only that the two product teams are working on it (Digital Trends, 2026).
Canceling a consumer app with 800,000 sign-ups the day before launch is an unusual move for a company Google's size. It suggests the AI Studio team concluded, late enough to scrap a finished build, that a separate app was the wrong surface for the feature set, and that consolidating inside Gemini mattered more than hitting the announced date.
What ties these together
None of these three events is about a new model capability. Each is a decision about price or product surface, made within a five-day span. OpenAI cut price to defend share against cheaper open-weight competition. DeepSeek shipped a model that made its own prior flagship look overpriced. Google canceled a nearly-finished consumer product rather than launch it on a second, separate app.
Taken together, they point to a market where the cost of inference is falling faster than most product roadmaps can adjust to, and where distribution decisions, not just model quality, are now being revised in real time.
What to watch next
Anthropic's introductory pricing on Claude Sonnet 5, $2 per million input tokens and $10 per million output tokens, is scheduled to expire August 31, 2026, with prices set to rise to $3 and $15 respectively on September 1 (Anthropic, 2026). Whether Anthropic holds that increase in a market where two of its largest competitors just cut prices is the next data point in this story.
Sources
- VentureBeat, "AI price wars: OpenAI cuts GPT-5.6 Luna prices by 80% as model competition shifts toward cost" - https://venturebeat.com/technology/ai-price-wars-openai-cuts-gpt-5-6-luna-prices-by-80-as-model-competition-shifts-toward-cost
- CNBC, "OpenAI cuts prices for two of its GPT-5.6 AI models as companies grow sensitive to costs" - https://www.cnbc.com/2026/07/30/open-ai-price-cut-gpt.html
- MarkTechPost, "DeepSeek Upgrades DeepSeek-V4-Flash-0731 with Major Agentic and Coding Gains" - https://www.marktechpost.com/2026/07/31/deepseek-upgrades-deepseek-v4-flash-0731-with-major-agentic-and-coding-gains/
- Hugging Face, "DeepSeek V4 Flash Is Now Official: What Changed in the 0731 Build" - https://huggingface.co/blog/ResterChed/deepseek-v4-flash-official-release
- Datanorth.ai, "DeepSeek-V4-Flash-0731: DeepSeek's new open MoE model" - https://datanorth.ai/news/deepseek-releases-deepseek-v4-flash-0731
- TechRepublic, "Google Got 800,000 Sign-Ups for an AI App, Then Canceled It" - https://www.techrepublic.com/article/news-google-cancels-ai-studio-mobile-app-gemini/
- Android Authority, "Google abruptly cancels its unreleased AI Studio mobile app" - https://www.androidauthority.com/google-ai-studio-mobile-app-canceled-3693645/
- Digital Trends, "Google just canceled its AI Studio mobile app, and it's not all bad news" - https://www.digitaltrends.com/cool-tech/google-just-canceled-its-ai-studio-mobile-app-and-its-not-all-bad-news/
- Anthropic, "Introducing Claude Sonnet 5" - https://www.anthropic.com/news/claude-sonnet-5