OpenAI launched GPT-5.6 final week and folded Codex right into a unified ChatGPT desktop app. Since then, the corporate has been operating flat out.
Early Tuesday, Tibo Sottiaux, engineering lead for Codex at OpenAI, mused on X that mixed energetic customers of Codex and ChatGPT Work may hit 8 million. The trajectory has been spectacular, particularly as the usage of Anthropic Claude has grown a lot in 2026. Codex had fewer than 1 million weekly energetic customers in February however hit 5 million by early June. Then GPT-5.6 launched on July 9, and the numbers accelerated sharply — 6 million by July 12, 7 million roughly 24 hours later, and eight million by Sunday.
That’s a progress curve most enterprise SaaS merchandise by no means see in a lifetime, not to mention in 5 months.
At 3:34 p.m. Japanese on Tuesday, Sottiaux posted to X, “Hi there. We now have reached 8M energetic customers throughout Codex and ChatGPT Work.” He additionally famous that OpenAI was resetting utilization limits for all customers. And round 2 a.m. Japanese on Wednesday, he up to date his followers on X with this progress stat: “Embarrassment of riches. However seems to be like we would hit 9M quickly. Ought to we reset the ChatGPT Work and Codex utilization once more or give it some area?”
What broke at launch
OpenAI merged the standalone Codex app into the ChatGPT desktop app, launched ChatGPT Work as a brand new agentic mode for data employees, and commenced sunsetting the Atlas browser — all in a single day.
Based on the corporate, demand surged virtually instantly, with visitors roughly doubling OpenAI’s earlier peak inside 48 hours. The sudden inflow uncovered a number of scaling points. In a detailed thread revealed on July 12, Sottiaux outlined the group’s response, which included optimizing inference to extend capability by about 10% per subscriber, lowering the context window from 372,000 to 272,000 tokens after the bigger restrict created unintended billing points, rolling again experimental reasoning-effort settings (internally often known as “juice” values), and patching overly aggressive multi-agent habits on the highest reasoning ranges.
OpenAI additionally quickly eliminated the five-hour utilization cap for Plus, Enterprise, and Professional subscribers — a transfer that quantities to probably the most beneficiant entry the product has supplied since launch.
Neighborhood reactions have been cut up between those that learn the context window rollback as a stealth downgrade and people who credited Sottiaux for explaining the operational trade-offs publicly. OpenAI CEO Sam Altman weighed in with what amounted to a positioning assertion towards rivals who, in his framing, deal with customers with contempt.
Rivals reply to the surge
Inside hours of OpenAI saying its 7 million person milestone, Anthropic prolonged its Claude Fable 5 promotional pricing by way of July 19 and bumped Claude Code’s weekly utilization limits by 50%. Whether or not one induced the opposite is unknowable, however the overlap offers us lots to take a position about.
Since Thursday, GPT-5.6 Sol has risen to second on Area’s agent leaderboard after 7,800 real-world agentic periods, and plenty of builders now see OpenAI because the chief in AI coding once more.
Price per job issues
Builders decide programs by the price of finishing a job. Cognition, for instance, reported that its Devin Fusion product, powered by Fable 5, may be cheaper per accomplished job than Anthropic’s pricier Opus 4.8 as a result of higher delegation reduces pointless work. In 81% of Fable-led runs, the lead mannequin by no means edits code. The takeaway is {that a} dearer mannequin can nonetheless decrease general prices if it avoids wasted work.
In 81% of Fable-led runs, the lead mannequin by no means edits code.
The harness is turning into the product
OpenAI’s determination to merge Codex, ChatGPT Work, and its built-in browser right into a single desktop app displays that shift. The identical is true of its plugin structure, which connects the app to Slack, Google Drive, SharePoint, CRMs, and calendars. OpenAI is popping it right into a workspace that sits atop the instruments individuals already use.
Anthropic has moved in the same route with Claude Cowork, which developed after the corporate noticed customers already utilizing Claude Code for excess of software program improvement.
Utilization caps stay the bottleneck
Proper now, the issue stays the utilization cap. Even with the five-hour restrict lifted, Codex and ChatGPT Work are pressured to share a single weekly pool. When you run a heavy Sol Extremely session with multi-agent orchestration, you’ll burn by way of that allowance extremely quick. (Sottiaux is conscious of the headache and says a repair is within the works).
Within the broader race, OpenAI has the sheer gravity of 900 million weekly customers and killer benchmark scores. Anthropic nonetheless holds an enormous quantity of developer goodwill due to premium code high quality and enterprise-friendly pricing. And for the DIY crowd, open-source stacks operating GLM 5.2 or Kimi K2.7 look higher than ever if you wish to commerce somewhat pace for knowledge management. Coding brokers and data work are multiplying quick, and we haven’t even hit the weekend but.
Coding brokers for coding and data work are multiplying quick, and we haven’t even hit the weekend but.
Replace, 9:30 a.m. Wednesday, July 15: This story has been up to date to replicate ChatGPT Work and Codex utilization person progress of almost 9 million.
YOUTUBE.COM/THENEWSTACK
Tech moves fast, don’t miss an episode. Subscribe to our YouTube
channel to stream all our podcasts, interviews, demos, and more.
