Just now, a piece of news instantly spread throughout the circle.The release time of GPT-6 has been greatly advanced, and it will be officially airborne as soon as this month!If the news is true, it means that the GPT-5.x series will end with GPT-5.6 and become the swan song.
It is reported that OpenAI made a very bold decision this time - completely abandoning the old "Spud" base with about 4T parameters and instead betting on a new, larger pre-training base.
Obviously, OpenAI’s intention is very direct: use this new base to compete head-on with Anthropic’s Fable 5.1, which will be released in a few weeks.
And just yesterday afternoon, good news came:GPT-5.6 is fully lifted! See you on Thursday!
Yesterday at noon, Axios exclusively broke the news: Officials have officially lifted the GPT-5.6 access restrictions on OpenAI.
Immediately afterwards, OpenAI officially announced on X: The flagship model GPT-5.6 Sol, together with Terra and Luna, will be officially released to global users this Thursday (July 9)!
The "Wang Zha" model, which has high hopes, has finally been lifted after a month of "limited government-enterprise special supply".
Ultraman shouted to the developers: Happy Building!
Forcing the palace, negotiating and lifting the ban late at night
On June 26, OpenAI quietly opened the preview of GPT-5.6, which was limited to a few people.
Because the AI Standards and Innovation Center under the U.S. Department of Commerce directly settled in OpenAI, and even forcibly retained OpenAI’s technical experts to conduct security reviews.
According to the latest AI executive order, the release of cutting-edge models must go through real-time case-by-case negotiations.
Now, why did the authorities suddenly relent?
Industry insiders revealed that OpenAI’s experts experienced numerous real attack stress tests and vulnerability fixes in Washington, and finally successfully proved that GPT-5.6 is OpenAI’s most robust security system to date.
In today's announcement, OpenAI emphasized: "We cooperate with the officials, but the short-term restrictions are to pave the way for wider availability. We never want official early access to become the norm, and the technology cannot be monopolized."
Now, the boots are on the ground and the official green light is given. The long-awaited global developer community is now full of expectations.
Three Arrows: How strong are Sol, Terra and Luna?
This time, the three new models have a clear division of labor.
GPT-5.6 Sol: the strongest flagship
Sol focuses on cutting-edge frontier capabilities, specializing in complex coding, cutting-edge science and network security tasks.
Its biggest feature is extreme reasoning.
Some people say that the capability improvement of GPT-5.6 is comparable to the leap of GPT-4/o1 level.
Early tester Max Weinbach commented excitedly: “The scariest thing about this model is that it never gives up! If you throw it into extreme reasoning mode, it will fight like crazy and never stop until the problem is completely solved.
Others say that GPT-5.6 is extremely persistent and can run for a whole day even without the /goal command.
GPT-5.6 Terra: The King of Cost-Effectiveness, the 5.5 Terminator
The performance is infinitely close to the previous super flagship GPT-5.5, but the cost is cut in half!
This is a work tool specially tailored for daily high-frequency work. Its appearance means that GPT-5.5 is out.
GPT-5.6 Luna: Lightning fast, ridiculously cheap
Luna focuses on extreme speed and extremely low cost, and is the first choice for high-frequency calls to agents.
What’s even more exaggerated is the seamless integration of the hardware ecosystem. It is reported that GPT-5.6 Sol will be directly launched on the Cerebras inference platform, and its inference speed has reached a breathtaking 750 tokens per second!
In high-intensity scenarios such as coding, biomedical research and development, and network attack and defense, this experience is an instant hit.
No wonder some developers who tested in advance lamented: "When I couldn't use version 5.6, I felt like I was going crazy."
Just today, the update log of Codex CLI version 0.143.0 was accidentally leaked: the Amazon Bedrock platform has quietly added the GPT-5.6 Three Musketeers and provided support for Max inference strength.
Judging from the configuration, this batch of models is most likely specially prepared for Max gear.
The first batch of actual test bosses have spoken.
Now, netizens on X who have obtained the qualification for actual testing in advance have begun to comment.
The well-known developer Theo sent several Xs in a row, blowing it up.
GPT-5.6 It solves all the pain points I had with 5.5. It is extremely persistent, even if you don't give it the /goal command, it can run on its own for a whole day! Its understanding and orchestration of subagents is superb. It made me use my computer a hundred times more often, and when the closed beta ended and I couldn't use 5.6 anymore, I felt like I was going crazy!
Magicpathai CEO Pietro Schirano also gave favorable comments.
It is no exaggeration to say that this is the best model I have ever used. It’s fast, creative, and they finally solved the front-end design problem! I haven't checked the code I wrote for two whole months!
Wharton Business School professor Ethan Mollick compared Sol to Fable.
Sol's abilities are similar to Fable's, but the physical feel is completely different. Fable likes to run at his own pace and complete tasks independently; while Sol is extremely fast and prefers to accompany you step by step and interact with you frequently. When I haven't figured out what I want, I'll use Sol to hit it back and forth; when I can clearly define the task, I'll use Fable for the long run. If I have a hell of a problem, I go to Sol Pro.
However, AI entrepreneur Matt Shumer sang a different tune. He said bluntly:
Sol is a great model, but in most tasks I tested, Fable is still better and has more Agent properties.
The leader of LMArena sets the ultimate tone: the wise owl and the deadly rottweiler
As a heavyweight player, Peter Gostev, the head of LMArena, also released the most hard-core in-depth comparison on the entire network.
He gave a very visually impactful metaphor: Fable 5 is like a "wise owl", while GPT-5.6-Sol is a "Rottweiler" - it will bite the throat of the problem and never let go until it is completely solved!
In Gostev’s view, there is a strong conflict between the underlying logic and body feeling of the two.
First, Fable 5 seems more arrogant.
Fable is inherently smarter and more insightful. But when building a new benchmark test, Fable showed "arrogance" - it ran in 40 minutes. Although it sounded smart, it was just based on feeling and scored 100 points on its own.
GPT-5.6-Sol was extremely diligent and took two full days to create a perfect benchmark that was fully tested and worked well.
In contrast, GPT-5.6-Sol is a work machine with a sense of security.
Although Fable still excels at hand-crafted UI and text writing from scratch, it always misses key features.
GPT-5.6-Sol brings an unparalleled sense of security - give it 8 to-do items, and it will complete them without compromise, without missing any items. with powerful/goalcommand, Sol can run continuously at high frequency for several days and nights.
In addition, Sol has achieved a qualitative leap in following existing code patterns, video editing (can seamlessly process 1-hour footage), managing sub-agent communications, in-depth research, and token efficiency.
Gostev even shared an extremely real "swear counter": in the GPT-5.5 era, the frequency of swearing at AI when he was writing code was 4-5%; after using 5.6-Sol, it dropped sharply to 1-2%; and once it was switched back to 5.5, the irritability index instantly soared to 7%!
In short, in terms of IQ alone, Sol is indeed inferior to Fable. But it's a tireless super-work machine that doesn't preach to you.
From this, Gostev gave the strongest workflow solution for current large models:
Use Fable for architecture discussions and brainstorming → leave it to GPT-5.6-Sol to frantically scratch the code and perform dirty work → finally use Fable to write documents and communicate externally.
Just tomorrow, the three models of GPT-5.6 will be officially launched.
This time, it’s the turn of developers around the world to get their hands dirty and score.