GPT-5.6 is FINALLY HERE (WOAH)
Video Overview & Insights
Check out GPT-5.6 coming to Box soon! https://bit.ly/4pkwci0
If GPT 5.6 can open the real desktop app and recreate its features over days, what kinds of enterprise workflows does that make possible that older models still can’t reliably handle?
GPT-5.6 Review site: https://signals.forwardfuture.com/gpt-5-6-review/
GPT-5.6 Relay: https://github.com/Forward-Future/gpt-5-6-relay
Yeah... the Minecraft clone is fun to watch. Whether capability holds up in sustained production coding is the question nobody asks in these demos.
Join My Newsletter for Regular AI Updates 👇🏼
https://forwardfuture.com
Nice, this is exactly where AI gets super useful for everyday workflow—have it break the big update into one quick action plan first, then e
My Links 🔗
👉🏻 X: https://x.com/matthewberman
It's spelled "whoa".
👉🏻 Forward Future X: https://x.com/forwardfuture
👉🏻 Instagram: https://www.instagram.com/matthewberman_ai
ask it what happens when you put a ball in an upside down cup and lift it off the table :D
👉🏻 Discord: https://discord.gg/evGThyRv
👉🏻 Spotify: https://open.spotify.com/show/6dBxDwxtHl1hpqHhfoXmy8
“This ran for five days” followed by “imagine how long it took to create Excel” is exactly what’s fucked about the AI-creator mindset. What an asinine thing to say out loud.
You didn’t recreate the achievement of building Excel. You told a model to clone a product that engineers already spent decades designing, testing and improving. “Excel clone” and “feature parity” already give it the product, features, UI and end goal.
Your perspective is completely out of touch with the human work it completely depends on.
Media/Sponsorship Inquiries ✅
https://bit.ly/44TC45V
A thoughtful review that goes beyond benchmarks and focuses on real-world performance. As AI models continue to improve, the biggest differentiator isn't just raw capability—it's reliability, reasoning quality, and how well they fit into everyday workflows. Practical evaluations like this help teams make smarter AI adoption decisions.
Links:
https://openai.com/index/gpt-5-6/
I’m kind of starting to hate all of these videos sucking off AI when it kind of still sucks. It can’t answer factual questions right. It can’t accurately comb through public domain literature to give correct answers. Godot has banned the use of AI because it’s writing so much slopcode that they can’t keep up with it. It’s just not super spectacular.
More User Perspectives
Pray to Jesus Christ and accept, he will speak to you Matthew.
@DocRogerThatWe have to be a little realistic here. That is 2% feature parity with excel and it took five days and $50,000 subsidized tokens. We aren't there yet. I could build that in 2 days on 3 ramen cups and a couple diet cokes.
@manawa3832New model comes out " omg this is the most advanced model ! "
@NachinBombin2Great, I'm going to ask Sol to go through my tax records and keep working until it finds loopholes saving me thousands.😁
@ElementaryWatson_fafowell it didnt realy shine in long term tasks... i reverted to 5.5 - this didint age well..
@dessmodeIt's more like comparing a maxed out Civic with a stock 5.0 V8 Mustang.
@SocialTestBedFable feels like a Ferrary? its trained on the same human data, uses twice as many tokens, like what did you expect to use double tokens and have worse results?
Clearly GPT is more efficient because its a better model, you can always increase token usage or make a more creative/divergent agent.
Making games with Fable and saying this game is definitively better because the program used Ivory yellow for the sky is more sophisticated is mental.
If you say Claude Fable is the best ("ferrari"), why would you still keep using the "suped up Honda Civic" ? Price alone should be the issue right? 🤔
@forretresssZero comment on grok 4.5 = credibility gone
@JesselivenomoreIt is so excellent that it deletes users' files, people reporting on X... 😅
@mklobucaricnow this is something else.
@off-gridlifestyleThat's a terrible analogy. TERRIBLE.
@justdrewit2219thumbs down for the thumbnail
@acejace00This is maybe 5% of excel capabilities.
@MichielvanderBlonkBox’s benchmark is so stupid 😂
@boiserunnerMatthew! I have a question! When you're running /goal on these super long running tasks, how do you handle compaction??? Do you just let the model hit 1M tokens, or do you handle compaction in a different way?
@BenLittle-wf6oqIt’s great for coding. Everything else is even more generic than any previous model. It prioritizes generic information over certain correct context information.
It somehow keeps getting more and more useless for anything other than a calculator with a dictionary or coding worker.
It is also so much harder in dumb generic and harmful guardrails that also ignore context.
Great to execute jobs and dumb enough to ignore any common sense of general knowledge or context knowledge to know if and how to proceed.
How much did it cost for running those demos? :)
@nileshacharya3691How many tokens did it take to produce that result? Sounds like the token is the new bitcoin. What's the backslash for?
Think I'll go ask another AI.
Just a note: you can unlock the "max" effort from your app's model settings, natively in the app. You're using Extra High which is a step below the actual smartest version. In some benchmarks, the jump between Xhigh and Max is large, so it's worth trying. Also, Ultra uses Max effort but with added adversarial subagents, that's pretty much all it is (which is cool in its own right).
@stateofthecatTell me how much money your making with ai.
@RightSideNewsSeriously ? You cut A 5 days run for Excel clone and it wasn't finished ? Same for Minecraft after 7 days ?
I call it failure !
It had to copy the actual Excel on the computer ? What a bad strategy ! scanning the online help and coding each functionality would make more sens to me.
I suspect you find it amazing because you were paid to say so.
Scam
@stevenpendragon8624It seems to have made Google Sheets 😂
@Olav3DWe need to use star citizen for the benchmark now ❤
@timmygilbert4102make acrobat pro clone for public?
@loc5706Do grok now
@OpalStaticbut how many tokens did it take?
@michaelswchanWhat's the point of these copies of existing applications? So what? You're not going to get better than the original: it's black box emulation. Matt's description of OpenAI's approach to clone Excel is literally doing that. Methinks your privileged all-you-can-eat token buffet demonstrations are a little out of touch with reality. This AI power is meant for original people to build stuff we haven't seen or imagined yet. Not more of the same ol', same ol'.
@japethstevens8473Matt - the burning question is, how many tokens burned over 5 days??????
@ai.positioningThere are nuances to Ultra that I don't think are conveyed exactly right. It's not just a matter of pushing harder on reasoning, it's doing more something like multi-agent orchestration, maybe synthesizing responses based on multiple opinionated agents for a slightly better response. The announcement states: Max gives GPT‑5.6 even more time than xhigh to reason and explore alternatives, run checks, and revise its approach. Ultra goes further by coordinating four agents in parallel by default, trading higher token use for stronger results and faster time-to-result on demanding tasks.
@michaelberg72015 days, 7 days. I can only imagine how much that cost.
@katie-cj5gyThis seems lack luster… 5 days to make an excel clone, congrats what a waste of time & tokens😂
@jamesrowdyDon't forget the Honda Acura NSX. I am already waiting for gpt 6 to get blown away all over again
@v0lrath1985Hi Matt. It looks like your YouTube channel is heavily infected by the Groklorum bot virus
@perhansen8170I think the Ferrari Legal team might want to talk they don't come optimised you say
@MBato89The f is with all those bots advertising "Groklorum"
@iven4843How much did it cost to get your Excel and Minecraft clones?
@myboringdesktopPlease drop the link for the GPT skill
@itachi2023How many tokens / costs per each run?
@Kulgan_EUDoes it have the Excel Flight Simulator?
@theoryofmine7473what is this fable youre talking about ? never touched it, never got the chance to work with it ...
@DanielSosig