Last week, OpenAI launched GPT-5.6 and the latest Codex/Work functions. Its excellent capabilities and cost-effectiveness have been very popular among users. CEO Sam Altman said on the X platform today (14th): The usage of its agent products (Codex and ChatGPT Work) has surged 2.5 times in the past week. At the same time, the price of the latest model GPT-5.6 Sol is only half that of its competitors, and the token efficiency is about twice as high, which is equivalent to only 1/4 of its competitors.
2.5x growth curve per week
Altman directly wrote in the post: “In the past week, the usage of our agent products (Codex and ChatGPT Work) has increased by 2.5 times. Welcome.” In just one sentence, there is a whole year of planning behind it. Since Codex gradually transformed from a coding assistant to an autonomous agent last year, OpenAI has been preparing for this breaking point. Now it seems that this moment has arrived.
2.5x increase in usage of our agentic products (codex and chatgpt work) in the last week!
welcome.
— Sam Altman (@sama) July 14, 2026
As early as April this year, OpenAI’s Codex had reached 3 million weekly active users. And the most terrifying thing is that the number of active users has been reached from 6 million to 8 million in just three days. This shows how popular the new version of gpt-5.6 and CodeX is (or the world has been suffering Anthropic for a long time).
Hello. We have reached 8M active users across Codex and ChatGPT Work.
We are once again resetting the usage limits for all. And we continue to not have the 5h rate limit as well, allowing everyone to explore the boundaries of GPT-5.6 Sol and discover how ambitious you can be.… https://t.co/pKunziOGRm
— Tibo (@thsottiaux) July 14, 2026
Internal adoption data shows that OpenAI not only sells agent products to customers, but also almost fully uses them among its own employees. By the end of June, internal data showed that 98% of OpenAI employees were using the Codex agent, and not just the engineering department, with more than 85% of employees using it on a weekly basis across departments. Even more astonishingly, Codex usage in research scenarios grew 56x in seven months, and the number of requests to the agent for eight hours of autonomous tasks increased 10x. When an AI company’s own employees rely heavily on its own agent products, such a growth curve is more than just marketing rhetoric. While most companies are still in the “trial AI agent” stage, OpenAI has already crossed the threshold from “trial” to “dependence”. This also means that OpenAI itself is the most demanding user when developing agent products, and its product iteration speed is naturally much faster than competitors that have no internal large-scale use experience.
GPT-5.6 Sol: Price cut in half, efficiency doubled
Altman revealed in another post on the same day that the newly released GPT-5.6 Sol model is only half the price of Claude Fable 5 on many tasks, and the token efficiency is about twice as high. “Happily delivering the same results for a quarter of the price.”
GPT-5.6 sol is half the price and ~twice as token efficient as fable in many cases for accomplishing the same task.
happy to deliver at one-quarter of the price.
— Sam Altman (@sama) July 14, 2026
According to the pricing announced by OpenAI, the GPT-5.6 series is launched in three versions: Sol (input $5 / output $30 per million tokens), Terra ($2.50 / $15) and Luna ($1 / $6). Among them, Sol’s benchmark score exceeds Fable 5 by about 11.4 points under the medium reasoning setting, but the cost is only about a quarter of it. Compared with Anthropic’s pricing strategy, Fable 5’s input price is about $10-15 per million tokens, and its output price is about $60-75. Sol’s price advantage is quite obvious. After conversion, the actual cost of using Sol may be only 20% to 30% of Fable 5.
Not to mention that OpenAI often resets the quota and gives away reset times when nothing happens, which is obviously more generous than Anthropic. Although the latter model is easy to use, its mysterious Token usage calculation mechanism and ultra-fast consumption have made developers complain. Originally, Anthropic was going to remove Fable 5 from the subscription system and change it to full API billing. However, because OpenAI is currently gaining too much momentum, the company has been extending the usage period and increasing the usage quota (laughs). I personally guess that it should not dare to really use Fable. 5 removed from the subscription system.
We’re extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.
— Claude (@claudeai) July 12, 2026
The sharp drop in price directly led to a surge in usage. It all started with the release of GPT-5.5 on April 23. This version was specially optimized for proxy capabilities and provided a more powerful engine for subsequent Codex and ChatGPT Work. A month later, GPT-5.6 was released in mid-May, further cutting the price in half, and the usage immediately increased sharply. Judging from the timeline, the entire cycle from model upgrade to price reduction to usage explosion took less than three months, which was much faster than outside expectations.
Conclusion
Codex was originally positioned as an “AI code assistant”, but in recent years OpenAI has gradually turned it into a more autonomous agent form. Codex is no longer just about completing the code, but can independently plan, execute, and debug hours-long development tasks.ChatGPT Work It extends similar capabilities to non-engineering scenarios, allowing ordinary workers to assign complex tasks through natural language, such as organizing reports, generating presentations, analyzing data, etc. Just like Claude Code made Anthropic explode, it can be seen that people don’t just want an AI that can answer questions, but an AI assistant that can actually do things. The result of the two giants competing for users is ultimately satisfying to us users, and I hope they continue to fight.
OpenAI 正式發表 ChatGPT Work：結合 Codex 與 GPT-5.6 的 AI 代理，跨應用程式自主工作數小時
Source: KOCPC Chinese