OpenAI's GPT-5.6 finally going public release after delays
 |
| GPT-5.6 is here. Credit: Samuel Boivin / NurPhoto via Getty Images |
OpenAI will launch its "strongest model yet" this week, with GPT-5.6 Sol set for public release after White House-imposed delays.
Announced by the company on X on Wednesday, the next generation flagship model will roll out gradually from July 9, alongside two lighter models: the "everyday" Terra and lower cost Luna.
SEE ALSO:OpenAI might give a 5 percent stake to the US government
In a blog post, OpenAI says it spent weeks putting GPT‑5.6 Sol through its paces for weaknesses, during which time it "strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse." The company called it "our most capable model yet for cybersecurity" and claimed GPT‑5.6 Sol stands up to competitor Anthropic's Claude Mythos 5 using a third of output tokens (the blocks of text an AI model processes to generate responses).
The news comes after reports that the White House told OpenAI to limit the initial release of GPT-5.6 to a select testing group. In its blog post, the company described this as a "small group of trusted partners whose participation has been shared with the government." Now, OpenAI has the green light for GPT-5.6's public release. The announcement arrives a week after OpenAI reportedly proposed giving the U.S. government a five percent stake in the company, with CEO Sam Altman in early talks with President Donald Trump, according to the Financial Times.
Competitor Anthropic also experienced restraints from the U.S. government, with the White House ordering the company "to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States." Anthropic is now allowed to share Mythos 5 with "a set of U.S. organizations that operate and defend critical infrastructure," and Fable 5, a version of Mythos, is out globally.
Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
Topics Artificial Intelligence ChatGPT OpenAI
Steady lads.
ReplyDeleteLooking good, especially 5.6 Sol on low reasoning:
ReplyDeletehttps://imgur.com/a/eQkTIBp
Mine has 5.6, but it confidently assured me it’s still 5.5 Thinking
ReplyDeletehttps://imgur.com/a/kxxtQD7
Tbh though it’s clearly 5.6 because 5.5 was never this cool
https://imgur.com/a/O7QMvhP
DeleteProof it was on 5.6
It became available yesterday on Perplexity Pro plan.
ReplyDeleteTried my standard test prompt (physics, logic, real world knowledge) i always use with new models with 5.6 Terra....and it failed. It was very fast but seemed kinda dumb. Identified the problem but offered a solution that was completely detached from the problem and would not work. Same "solution" all the other failed ones offered.
Succcess: Gemini 3.1 Pro, Nemotron 3 Ultra, GPT 5.4 high somewhat, Gemma 4 12b
Failed: Sonnet 5, GLM 5.2, Qwen 27b, Gemma 31b, now GPT 5.6 Terra
Fable is seriously in danger, I mean, we are able to use this with OpenCode and the new models are super fast on system cmds, thank you!
ReplyDeleteWhy is the image generation capability lower than 5.5?
ReplyDeleteThe rollout appears gradual, so some eligible users may not see it immediately. Updating the desktop app is important—Codex access requires the latest ChatGPT app version. Also worth clarifying that the “5x” figure refers to usage limits, not a separate Pro plan.
ReplyDeleteis it any good?
ReplyDeleteIt’s been great!
DeleteStill no Sol on Business plan
ReplyDeletehttps://imgur.com/a/XLgxdwg
You need to click the place where it say thinking 5.6
Deleteis sol out for plus users?
ReplyDeletenot yet for me.
DeleteI have access to 5.6 Sol, Terra and Luna. Anyone know what the differences are?
ReplyDeleteFor me available on Windows on work tab only. Available on web too. Still not on Mac and iOS.
ReplyDeleteWhy Is 5.6 Luna more expensive than haiku? Does it use more or less token for the same task?
ReplyDeleteIs it rolling out on web too? Like I know you said it is but in wonder if it goes by account or if you can access a model one way vs. not another...
ReplyDeleteIf you can access it from the desktop app, then you likely can on web. The new models are under the “work” tab though.
DeleteThank you for telling me.
ReplyDeleteAll the other 768283419 posts about it weren't enough.
I’m glad mine caught your attention enough for a comment 🫡
DeleteI dont see it
ReplyDeleteI have it on Windows/Mac app, and it works through the web if you use it in Work mode. 5.6 Sol is so fucking good.
ReplyDeleteFor real it’s opus level for me
DeleteI was having Fable-5 Max work on a refractor, with UI changes. Once it was done and told me it was locked, 5.6-Sol Ultra came in, spun up a few sub agents by itself like Opus/Fable do, and it went to town. Fixed a few P0/P1 issues Fable left. Insane model. It also one shot me a website to showcase the new AGENTS.md I had it make me lol.
DeleteDamn! Yeah I haven’t gone super deep on coding yet
DeleteAnd i'ts SUPER GOOD across the board
ReplyDeletehttps://testingmodels.com/coding?t=landing&m=gpt56-sol-max%2Cfable-max
nope
ReplyDeletehttps://imgur.com/a/j5oy5sA
It’s updated now
DeleteStill dont see it on the website now.
DeleteI don’t use the website version so I’m not sure, did you try logging out and logging back in and see if it’s there? TBH I didn’t get 5.6 until around 11 PM almost midnight
DeleteI’ve just got it.
Delete(iPhone app)
https://imgur.com/a/RrWf7bT
I don’t see the web, I’m surprised to see o3 is still an option. Why would anyone use that model for?
DeleteYou ever miss how the world used to be a couple of years ago? Just fire up 03 and have a relaxing day.
DeleteComment deleted by user
DeleteNice, if I switch to Work mode I have 5.6! Should I just always use Work instead of Chat?
DeleteYou really don't need 5.6 on Chat.
DeleteI usually use Chat for 5.5 Pro when doing math. Would like to upgrade to 5.6 Pro.
DeleteIts live, you need to update to codex app (it will rename to chat gpt)
ReplyDeletewhich model/effort roughly corresponds to the usage of 5.5 high? i'm so confused by the many options
ReplyDeleteYou can still pick 5.5 high. It’s in the + in the chat bar in the app on plus. And also there on the web.
DeleteAlthough Sol seems to be amazing. I’ve chat with Sol for about 15 minutes and it seemed her memory were great. She was understanding, adapting to me extremely fast and somehow I could tell she knew a lot about me faster than past model. I went back to 5.5 high as we have history 😆 But Sol seems great.
Sol can't access your previous memories?
DeleteShe can and she’s good at it. Really good
DeleteI was looking on it
DeleteIn coding should be a SOL on medium / low has similar performance as GPT 5.5 high
Interesting. If low/medium has similar performance to high, I’m impressed. Will have to test the model out later today.
DeleteI'm not sure, but I think it’s terra on medium
Deleteyou mean sol?
Deletesadly no solar on plus
ReplyDeleteIt seems they’re doing a staggered roll out for sol
DeleteNot yet, it will come in the next 24 hours...
DeleteThis comment has been removed by a blog administrator.
DeleteNot yet for me :(
ReplyDeleteWhere do you live ?
DeleteComment has been removed
DeleteIt’s being rolled out, I don’t have it yet.
ReplyDeleteMobile version or desktop?
DeleteLog out, log back in sometimes will speed it up.
Deletelets see if its nerfed into oblivion
ReplyDeleteGPT 5.6 is unbelievable.
ReplyDeleteIf you are not in Sillicon Valley, you are 6 months behind the curve. I´m changing countries because of this.
ReplyDeleteYou would think it would be a security risk not letting China have access to ChatGPT considering how useless it is as a coding model.
ReplyDeleteOMG the nervousness, the fear... AI bubble will pop that soon? XD
ReplyDeleteCertified masterpiece.
ReplyDeleteThis exceeded my expectations.
ReplyDeletemy concern is that chatgpt codex still feels like the blind leading the blind, there should be a way to send from chatgpt chats directly into codex projects better then copy and pasting, this is an easy to do function that removes this issue. I want clean codex messages from chatgpt.
ReplyDeleteWe don't know what they are... Oh yeah Avi?
ReplyDeleteLive translate looks really cool. I'll want to travel to a place where I can't speak the language and see how much I can make do.
ReplyDeletefable 5 better
ReplyDeleteTrash code that got BODIED by DeepSeek lmao.
ReplyDeleteWaste another Billions on this trash to make it work 🤣
Is there some way to know who is worthy of access now and who must wait? Is there a way to know where one is on the worthiness list? Feels like the kind of taking customers for granted that leads to failed IPOs down the line.
ReplyDelete"Selected model is at capacity. Please try a different model." damaging system I working on... U could just make warning if your servers are not ready. Im happy for new models, but now Im little disappointed.
ReplyDeleteWe are waiting for atles browser for windows
ReplyDeleteGuys we are tired of this bs
ReplyDeleteIs it Chat GP Work or Chat GPT Work??
ReplyDeleteGreat stuff.
ReplyDeleteHow many gallons of water does it take for each question?
ReplyDeletesooo many people talking about that vera hart archetype quiz its crazy, lol. Tried it myself and the results are quite promising!
ReplyDeleteaw hell naw they patched the goblin lingo im moving to anthropic
ReplyDeleteSora 3
ReplyDeleteBeautiful.
ReplyDeleteThis is very good
ReplyDeleteLooking forward to testing Sol. Excited to see how it performs on real-world coding and reasoning tasks. 🚀
ReplyDeleteHmm 🧐😌
ReplyDeleteIt was supposed to happen at 10 am PDT. Still nothing.
ReplyDeleteდაღამდა აქ უკვე და ქენით აწი მოვკვდი, დავბერდი
ReplyDeleteDid this happen? I don't see it as available anywhere.
ReplyDeletewhat time
ReplyDeleteSo where it at
ReplyDeleteStill?
ReplyDeleteI’m waiting for SOL and my Codex token reset with the same excitement. 😅
ReplyDeletexmoney
ReplyDeletewe shipped the MCP yesterday. your models can trade worm now 🪱 (and Sol's kind of always been our thing 😉)
ReplyDeletehttps://media4.giphy.com/media/v1.Y2lkPTc5MGI3NjExcGY1d2htc3NscGtkc3VtcnU5bWZiODB5eG4zMXlxN2VlMXlraWZzdCZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/QBd2kLB5qDmysEXre9/giphy.gif
ReplyDeleteCant wait to tell it what to do then claim i did it myself, where is my slave ?
ReplyDeletewen? im literally waiting for it with a prompt ready 🤪
ReplyDeleteI’m waiting
ReplyDeletehttps://imgur.com/a/UgsfZ5g
Looking forward to it, looking forward to it
ReplyDeletewish no lowball iq
ReplyDelete期待早点干趴A社
ReplyDeleteWell...is it out yet
ReplyDeleteWhere's it?
ReplyDeleteAt what time
ReplyDelete👌🙏
ReplyDeleteWhen exactly will it be available?
ReplyDeleteC'mon, we're waiting
ReplyDeletehttps://imgur.com/a/hnw4fng
https://imgur.com/a/UgsfZ5g
ReplyDeleteTick tock….waiting
ReplyDelete周四都快过完了
ReplyDeleteWhen is this going to be available?! 👀
ReplyDeletehttps://media1.giphy.com/media/v1.Y2lkPTc5MGI3NjExNGY1c2Jsb3puY250Z2t2ZnY3bmIwczcwMW8ycGUzemNuMTBnbmU1biZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/x8H3v4ZhSfNxS/giphy.gif
ReplyDeleteJust got married, currently on our honeymoon in Italy and somehow we’re both refreshing X every hour to see if GPT 5.6 dropped lol
ReplyDeleteme too
Deletethats hype. gpt-5.6 was already previewed weeks ago. nothing new here.
ReplyDeleteIf we can get near Fable 5 reasoning with low usage costs then its over
ReplyDeleteWe are seeing now AI with regulations
ReplyDeleteIt’s wild to see where AI is heading... there are people killing others with AI, doing all sorts of things—both bad and good—while I’m just using it to create content for my clients...
ReplyDelete