OpenAI cuts GPT-5.6 prices and puts Codex inside ChatGPT for work

OpenAI made cost the headline with steep price cuts to its GPT-5.6 family: Terra dropped 20% to $2 per million input and $12 per million output tokens, and Luna fell 80% to just $0.20/M input and $1.20/M output, while flagship Sol pricing stayed put. The moves, reported by CNBC, directly answer the deflationary pressure from DeepSeek and other cheap-inference challengers, effectively conceding that mid-tier frontier intelligence is becoming a commodity.

Alongside pricing, OpenAI repositioned ChatGPT itself. With Codex now built in, the company is pushing ChatGPT 'beyond answering questions to real work' — over 5 million people use Codex weekly, and President Greg Brockman declared 'chatgpt is becoming an agentic browser.' Independent developer Simon Willison called Luna 'a bit of a beast,' testing it in Datasette Agent after the 80% cut and praising how 'furiously quick' it generated SQL, HTML, and JavaScript.

The strategic logic is volume over margin: cheaper tokens plus agentic capability expand the addressable surface for ChatGPT as a work platform, deepening lock-in through Codex-driven workflows rather than raw model access. Sam Altman amplified consumer use cases too, describing a ChatGPT-generated morning podcast built from family calendars.

OpenAI also opened a ChatGPT for Academic Researchers program to 100,000 scientists, extending its enterprise-and-institution land-grab. The competitive read: with Luna now at $0.20/M input, OpenAI is undercutting much of the mid-tier market while betting its agentic ChatGPT experience — not price — becomes the moat. The risk is margin compression across the industry, and whether 'agentic browser' reliability keeps pace with the marketing as more users delegate real tasks.