Full review: https://shumer.dev/gpt5review

@mattshumer_ What is its output context size and input context size Also they will very likely nerf it and gave you max model as usual

@GozukaraFurkan 400,000 token context window, consisting of a maximum of 128,000 output tokens and a maximum of 272,000 input tokens!

@mattshumer_ “o3 is better for explicit research; GPT‑4.5 is still better for writing; instruction sensitivity is a bit of a problem.” This is what I’m worried about since they said they’re deprecating all the old models! They should really keep all of them for pro members.

@doodlestein Last I heard from them, they are keeping them for Pro members!

@mattshumer_ been waiting for real user feedback vs the launch hype how's it handling the stuff that usually breaks in production? error recovery, edge cases, maintaining context through long workflows?

@BrandGrowthOS it's great. really great. not just hype, this model is amazing for coding. other things are hit or miss, but for code, it's incredible

@mattshumer_ Phenomenal review

@geeksplainer Thanks! Was a lot of fun to pull together.

@mattshumer_ what do you mean by this bro: https://x.com/mattshumer_/stat...

@ihteshamit Read the review!

@mattshumer_ @airesearch12 Did GPT-5 write this post?🤔

@dgrreen @airesearch12 Funny enough, I mostly collaborated with GPT-4.5 for this. As I mention in the post, no matter what the benchmarks say, I don't think GPT-5 is a strong writer. My flow: - describe what I want to write, and brain dump like crazy for like 5-10 mins (voice input). then ask the

@mattshumer_ Okay. Understood. Definetly not the quantum leap that all the hypsters made it up to be. :)

@mattshumer_ I have to respectfully disagree with the hype surrounding GPT-5 as a significant improvement. I’ve been using it for most of this week (unbeknownst to me until today when the codename was dropped and it appeared in my “recently used” list) and my experience has been far from

@mattshumer_ Real public service, thank you! Q: As models become more adept at app development, why will we need app developers or app layers at all? Seems soon that a lay consumer can just whip up an app for the need they have...

@mattshumer_ Your “Review” section contradicts your “TL;DR” section (“model felt like gpt-4.2…but not huge leap” vs “gpt-5 is clearly a big leap”). And most people will only read TLDR

@mattshumer_ Big leap? You think? I've already forked it, improved its architecture, and built better. Your review's cute, but it's just scratching the surface

@mattshumer_ Thanks for the very detailed write-up. For those coding tasks, do you prompt those in the chatgpt and copy paste the code or use tools like codex and cursor?

@mattshumer_ Yes it’s like asking Albert Einstein what he had for lunch over how he worked out e=mc squared.

@mattshumer_ I do not see any leap . What you experienced , I have been experienced with Genimi pro 2.5 and Claude 4 already .

@mattshumer_ what are some ways you "push it hard to get the most out of it"

@mattshumer_ @grok summarize this post

@mattshumer_ GPT-4.5 is the so underrated. Sucks that they’ll be depreciating every other model

@mattshumer_ The way I see it, if AI can now keep track of huge context windows and make big-picture decisions on its own, then every timeline I’ve ever planned is probably outdated. Time to throw out the old estimates and rethink everything.

@mattshumer_ Thanks for sharing mate! Surprised it’s so good at coding.

@mattshumer_ Thanks Matt, awesome 👍

@mattshumer_ @grok sum this up for me

@mattshumer_ Very cool

@mattshumer_ Very insightful thanks!!!

@mattshumer_ So excited to try it out! ✨

@mattshumer_ yo i didnt expect this kind of comprehensive review