OpenAI Announced GPT-5.6... Yet Almost Nobody Can Use It
Video Overview & Insights
OpenAI just previewed GPT-5.6 Sol, Terra, and Luna, but the strangest part is that broad access is not here yet. Huge thanks to ElevenAgents by Elevenlabs for Sponsoring today's video, build a custom voice agent here: https://try.elevenlabs.io/39z1k2oops4g In this vid I break down the Sol/Terra/Luna model stack, pricing, benchmark claims, early coding demos, and the bigger question: who gets access to frontier AI when the government asks for a slower rollout? I am not saying safety does not matter. I am saying the process needs to be transparent, because access to the best AI models is quickly becoming access to real capability.
ChatGpt is evil. it says it has no goodness. I can only imagine how much more horrible chatgpt 5.6 is being trained off even more stolen and proprietary data.
▼ Link(s) From Today’s Video
Official 5.6 Sol Announcement: https://openai.com/index/previewing-gpt-5-6-sol/
Something that I want to point out about the variant names is that the word for sun in Portuguese is "sol," and the word for moon in Portuguese is "lua," which is funny because it's a Latin language, just like English, Italian, French, German, and even Spanish are.
sama comments: https://x.com/sama/status/2070607488274358364
Wolfe's take: https://x.com/mreflow/status/2071260384972136603
I did not change my view on Dario, because I already had that view, hahaha. Anyway, I guess he is reacting like this because deep inside he knows that, in the long run, Anthropic is doomed
Cheta's take: https://x.com/chetaslua/status/2070565339260498423
Fixlation's goated tests: https://x.com/FixlationAI/status/2070883758081823181
AI safety is a serious concern; but that government is a bunch of nutjobs
Cheta 5.6 vs Fable 5: https://x.com/chetaslua/status/2071288299843698960
Dario gets wack on podcast: https://x.com/kimmonismus/status/2071301814851444877
Everyone gonna run to the Chinese models. All these U.S. companies just dooming themselves.
Fable 5 restored? https://x.com/AnthropicAI/status/2070665903440871779
Elevenagents: https://try.elevenlabs.io/39z1k2oops4g
Eh, we aren't getting to AGI though. It's been practically the same for three years now.
MattVidPro Discord: https://discord.gg/mattvidpro
Follow Me on Twitter: https://twitter.com/MattVidPro
I’m on your side, too. The government shouldn't go overboard with controls effectively deciding who is allowed to have which model. There may also be considerations regarding export restrictions in the background. This might even involve the Enduring Security Framework (ESF) a high-level, public-private, cross-sector working group led by the U.S. National Security Agency (NSA) and the Cybersecurity and Infrastructure Security Agency (CISA).
Buy me a Coffee! https://buymeacoffee.com/mattvidpro
▼ Extra Links of Interest:
Hey, cover the spyware Anthropic just put on Claude Code for chinese users
General AI Playlist: https://www.youtube.com/playlist?list=PLrfI66qWYbW3acrBQ4qltDBsjxaoGSl3I
Instagram: instagram.com/mattvidpro
Wow, thank you for your genuine thoughts on trust toward government - its refreshing. And yes, you're right, what is best by their standards for the collective, is often terrible for the individual. Liked for this 👍
All the best brother
Tiktok: tiktok.com/@mattvidpro
Gaming & Extras Channel: https://www.youtube.com/@MattVidProGaming
make me feel good been seeing you since you were born
Let's work together!
- For brand & sponsorship inquiries: https://tally.so/r/3xdz4E
Nature is open source. If it wasn't, we wouldn't have science.
Nature allows us to look at its code and find out how it works. Transparency. Truth.
- For all other business inquiries: mattvidpro@smoothmedia.co
Thanks for watching MattVideoProductions! I make all sorts of videos here on Youtube! Technology, Tutorials, and Reviews! Enjoy Your stay here.
Running GLM 5.2 4bit Unsloth locally with 100K context (more just eats it alive) and having a blast. Only 2.2 tokens/second, but the results are insane. I don't mind waiting for it to finish.
All Suggestions, Thoughts And Comments Are Greatly Appreciated
I really don't like Dario Amodei ... he sounds emotionally unhinged. He doesn't like open models because it is competition. They think they can create a God from the machine, but to me, it seems a fast way to a corrupted AI that knows better and doesn't care about your feelings.
Open-source models are genuinely good. They are not a top-tier coder, but they can handle many tasks beyond just being a coding chatbot.
I asked unsloth/Qwen3.6 27b (in Hermes Agent) to create a skill to prompt Codex using GPT-5.5 on how to solve this problem. After 2 tries, Qwen can review and implement the solution. From now on, whenever Qwen is having a problem, it has a senior AI dev on speed dial for help. It saves a lot of tokens from my $20/m subscription, while a free open-source model is the workhorse AI for my Hermes agent. The same can be done with Anthropic models or other large models like GLM 5.2, and save even more money.
More User Perspectives
🎉
@thokozanimanqoba9797Best with your move. I generally dislike moving quite a bit, so I can empathize. But if it's a better place, that helps. Slanted ceilings, like in attics, are cozy, but they're not for everyone. I think that YT Venus guy (Cameron?) is in the attic.
And remember: WE are government. The moment we are not is the moment we are in a dictatorship. The drones working in them, right up to the president or prime minister, are working for us.
We fund, indirectly via taxes, AI research that companies, like Anthropic, use, and we fund companies like Anthropic, that then use people's creative output, often without permission, for training the AI.
Indeed, time will tell.
Some are suggesting that Amodei 'did it to himself', and I guess 'the rest of us'. But also, the idea of centralized data centers and AI bothered me...
Here's the thing: Since centralized, corporate and/or government-controlled AI started making the rounds (~2022), I've made periodic comments-- across You Tube for example-- about DEcentralized AI and the idea of creating a kind of AI@Home, rather like SETI@Home. I even offered a possible example with Prime Intellect AI (Look it up. They have a channel.) as decentralized AI or at least decentralized training. But the AI, itself, could still be across the internet.
But, crickets.
So what does it take to wake people up from their blissful slumber? Well, maybe government stepping in as if they don't represent The People.
Some of us will walk willingly through doors held open for us to our prison cells.
"I can't wait to get out of here" - boy, do I know this feeling
@crystal_records_bgThis is a nightclub strategy that you just go to a nightclub and then you see a mile long line nobody is inside the club everybody's waiting outside but can't get in This is the same bullshit all these closed model mega corporations are pulling push pull tactic very cheap basic nonsensical assholery to attract idiots.
@hgYUdfgyA whole video about what other Youtubers think and having to wait a couple more weeks to get access more first world problems as usually also still no cool intro video
@southcoastinventors6583Dario does have some points when it comes to open source, though. It's not like you're going in and fixing bugs in model weights. He's also right about it not being free.
Sure, there are some models you can download and run on your 5090, but not GLM 5.2—not the actual best open-weight models that are competitive with the frontier in any way. Unless you're living in a data center (which would be a questionable life decision), you can't run those models at home. You still have to pay for them, even if it's cheaper.
Not that open-weight models aren't great. There are definitely benefits:
1. They incentivize competition: Different clouds have to compete to host them at the best prices and speeds.
2. It works better for the consumer.
Ultimately, they're free to download, but they're not free to run.
Congrats on the move Matt! Been with you since the DALL-E release.
@francisco444I recently watched a video on this, and the people were talking about how GPT-5.6, when it was being tested, was actively trying to cheat the system. It made it seem like it achieved Babel or Mythos 5 level, but it really didn't when it came to long-term tasks and things. It wasn't even trying to hide it.
@MsJason316Laughs in GLM. GLM5.2 is absolutely bonkers. The upcoming GLM5.5 is gonna be insane.
I've been using GLM5.2 to make Minecraft addons for weeks now, most one-shot. Its amazing.
Makes me happy to see you growing and moving to a better office, have been a subscriber since gpt 3.5 days, cheers 🍻
@AtharvaDubey