OpenAI Drops GPT-5.6 Trio and ChatGPT Work in One Day: One-Line Slide Decks, Aimed at Claude's Heart
The month-long hold on GPT-5.6 finally lifted — and the same day, OpenAI shipped an agent that builds your slide decks for you.
Three Key Takeaways
GPT-5.6 hit general availability on July 9 in three sizes at once: Sol, Terra, and Luna. All three share a 1-million-token context window, 128K max output, and a February 16, 2026 knowledge cutoff. On price, the smallest, Luna, runs just $1/$6 per million input/output tokens, Terra is $2.50/$15, and flagship Sol is $5/$30 — Luna costs a fifth of Sol yet trails it by only 3.3 points. In command-line automation on TerminalBench 2.1, Sol hit a record 91.91% with its new ultra thinking mode. (For comparison: Claude Opus is $5/$25, Fable 5 is $10/$50.)
The same day, OpenAI also launched an agent called ChatGPT Work. It fuses the chatbot with the Codex coding tool so non-coders can spin up documents, presentations, and websites from a single prompt — all running on GPT-5.6. It went live July 9 on web and mobile for Pro, Enterprise, and Edu users, expanding to Plus and Business over the following days. As product manager Ty Geri put it: apply the model's coding ability to solve problems across every industry.
This is aimed squarely at Anthropic. ChatGPT Work is all but a carbon copy of Claude Cowork, which Anthropic shipped in January and which autonomously plans and executes multi-step tasks. GPT-5.6 was delayed last month over U.S. government national-security concerns; now unshackled, it lands right as both labs prep for public offerings and fight over enterprise accounts.
WangDou's Take
Two launch events in one day looks like a tech festival, but it's really OpenAI on the back foot. Claude Cowork shipped in January and grabbed a chunk of enterprise deals over six months; then GPT-5.6 got pinned down by the government for a month. So the first move after the ban lifts isn't to flex benchmarks — it's to stuff Codex into ChatGPT and turn it into a white-collar drone that makes your slides. Because companies don't care that you scored 91 on TerminalBench; they care whether they can hire two fewer interns to build decks. The pricing is the real knife: Luna at $1 per million tokens is five times cheaper than Claude Opus, a naked bid to pull small businesses out of Claude's bowl. And "grab enterprise deals before the IPO" just means both sides are racing to dress up their financials so they can sell high when they list. Models are turning into utilities — cheapest wins — and this fight comes down not to who's smarter, but to power bills and gross margin.
Source: Reuters/KFGO · Simon Willison · Engadget