“Okay so I didn’t super expect the results of the GPT4 vs. GPT4.5 poll from earlier today 😅, of this thread: https://x.com/karpathy/status/1895337579589079434
“@karpathy Nah this is awkward, not “high taste”, sorry Andrej :p https://x.com/jeremyphoward/status/1895354868342366648
“ChatGPT for macOS can now edit code directly in IDEs. Available to Plus, Pro, and Team users. https://x.com/OpenAIDevs/status/1897700857833193955
“!! https://x.com/sama/status/1896651354648818121
“ok fine maybe we’ll do a social app” / X https://x.com/sama/status/1895230925753233763
“once in awhile the high taste testers are right” / X https://x.com/sama/status/1896649668387541128
“@karpathy Ok, according to the system card (https://t.co/NtugtBMkTD), it’s trained “using new supervision techniques” https://x.com/rasbt/status/1895511885950357888
“Damn lol must have some high, or low, taste people testing here idk https://x.com/Teknium1/status/1895348781367140708
“I think this is the most insane part of 4.5 release to me. The knowledge cutoff is 2023 still. How do you even have a current pretraining run that didnt see data past 2023? So many API’s and libraries from there are now deprecated, and so many new ones created.. Did chatgpt 3.5” / X https://x.com/Teknium1/status/1895380611764015342
OpenAI Developers on X: “ChatGPT for macOS can now edit code directly in IDEs. Available to Plus, Pro, and Team users. https://t.co/WPB2RMP0tj” / X
https://x.com/OpenAIDevs/status/1897700857833193955
“@karpathy now the pricing makes even less sense..” / X https://x.com/Yuchenj_UW/status/1895338027041579269
“Think Deeper just got smarter. Now powered by o3-mini-high free in Copilot. That means fresher insights, more relevant answers, and AI that keeps up with the world, right when you need it. https://x.com/yusuf_i_mehdi/status/1897783236354515420
“my new favorite coding workflow: gpt-4.5 for brainstorming and planning claude 3.7 sonnet for building windsurf for all the agentic stuff watch the video for a quick example https://x.com/omarsar0/status/1896620019053895842
“SoS1 O1 and R1-Like Reasoning LLMs are Sum-of-Square Solvers https://x.com/_akhaliq/status/1896395391014433147
“Introducing Sora, our text-to-video model. Sora can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions. https://x.com/OpenAI/status/1758192957386342435
OpenAI reportedly plans to charge up to $20,000 a month for specialized AI ‘agents’ | TechCrunch https://techcrunch.com/2025/03/05/openai-reportedly-plans-to-charge-up-to-20000-a-month-for-specialized-ai-agents/
“an idea for paid plans: your $20 plus subscription converts to credits you can use across features like deep research, o1, gpt-4.5, sora, etc. no fixed limits per feature and you choose what you want; if you run out of credits you can buy more. what do you think? good/bad?” / X https://x.com/sama/status/1897036361506689206
“In conversations with leaders in professional service roles, the key line that OpenAI’s Deep Research crossed for a set of valuable work is that it now takes far less time & far less expensive personnel to check that the report is right (mostly it is) than to have humans write it” / X https://x.com/emollick/status/1897422975005286536
“developers can get started in a few lines of code pip install –upgrade “ai-gradio[openrouter]” import gradio as gr import ai_gradio gr.load( name=’openrouter:openai/gpt-4.5-preview’, src=ai_gradio.registry, coder=True, ).launch() github: https://x.com/_akhaliq/status/1895352428591227397
“My posts about gpt-4.5 have some interesting comments like you don’t know what you’re talking about, or you aren’t using it right or you’re a slop enjoyer etc. No no you don’t get it, you don’t train the *largest* model to be a model about “taste”- it needs to make me more” / X https://x.com/abacaj/status/1895516638704754727
“retracting my positive comments about gpt-4.5, don’t want to be seen as a low-taste tester https://x.com/vikhyatk/status/1896350962018869510
“GPT-4.5 — a model trained at the next level of scale:” / X https://x.com/gdb/status/1895225079333875781
“I think OpenAI missed a bit of an opportunity to show GPT-4.5’s strengths, to their detriment & to the AI industry as a whole by only using the same coding & test benchmarks when critical thinking & ideation are key AI use cases where 4.5 is good. Those are actually measurable” / X https://x.com/emollick/status/1895350182713462921
“@dwarkesh_sp There is OpenCanvas from @LangChainAI which is similar to OpenAI’s one but works with every model https://x.com/_philschmid/status/1897405585118912618
“OpenAI released GPT-4.5, its largest model to date, but one that lacks the reasoning capabilities of recent models like o1 and o3. Learn more in The Batch: https://x.com/DeepLearningAI/status/1898086241859440934
“How @ConsensusNLP uses GPT-4.5 for nuanced scientific and medical written analysis, and Structured Outputs to visualize levels of agreement across research papers: https://x.com/OpenAIDevs/status/1895531723640729907
“@lmarena_ai @xai lol…. these AI timelines are ridiculous. both gpt-4.5 and grok 3 are fun models to use” / X https://x.com/omarsar0/status/1896676260312670589
“BREAKING News: @OpenAI’s GPT-4.5 now tops the Arena leaderboard! With over 3k votes, GPT-4.5 landed #1 across ALL categories, and singularly #1 under Style Control / Multi-Turn 🥇 Huge congratulations to @OpenAI on this impressive milestone! 🙌 View below for more insights on https://x.com/lmarena_ai/status/1896590146465579105
“GPT-4.5 topped all categories across the board, with a clear leadership in Multi-Turn. 🥇 Multi-Turn 💠 Hard Prompts 💠 Coding 💠 Math 💠 Creative Writing 💠 Instruction Following 💠 Longer Query https://x.com/lmarena_ai/status/1896590150718922829
“GPT 4.5 + interactive comparison 🙂 Today marks the release of GPT4.5 by OpenAI. I’ve been looking forward to this for ~2 years, ever since GPT4 was released, because this release offers a qualitative measurement of the slope of improvement you get out of scaling pretraining” / X https://x.com/karpathy/status/1895213020982472863
“developers can get started in a few lines of code pip install –upgrade “ai-gradio[openrouter]” import gradio as gr import ai_gradio gr.load( name=’openrouter:openai/gpt-4.5-preview’, src=ai_gradio.registry, coder=True, ).launch() github:” / X https://x.com/_akhaliq/status/1895488615586607609
Introducing NextGenAI | OpenAI https://openai.com/index/introducing-nextgenai/
“DeepSeek R1 is joint #1 with GPT 4.5 on hard prompts with style control. Congrats to OpenAI team.” / X https://x.com/teortaxesTex/status/1896591303150104784
“We’re excited to finally share what we’ve been up to 🙂 Announcing Endex: An AI Financial Analyst Today we’re coming out of stealth and announcing our partnership with @OpenAI 🧵 https://x.com/TarunAmasa/status/1895166850742259922
“it’s very hard to get the math and ML right on a run as big as GPT-4.5, and requires difficult work at the intersection of ML and systems. @ColinWei11 , Yujia Jin, and @MikhailPavlov5 did excellent work to make this happen!” / X https://x.com/sama/status/1895490123690922445
“Looking forward to GPT-4.75” / X https://x.com/bilawalsidhu/status/1895634676557234548
“Excited to share more about @FactoryAI’s partnership with @OpenAI. The software of the future will be built by humans and AI, together, in one platform. Big thanks to @shyamalanadkat @edwinarbus and @OpenAIDevs for putting this together!” / X https://x.com/matanSF/status/1897694460592754829
“we are likely going to roll out GPT-4.5 to the plus tier over a few days. there is no perfect way to do this; we wanted to do it for everyone tomorrow, but it would have meant we had to launch with a very low rate limit. we think people are gonna use this a lot and love it.” / X https://x.com/sama/status/1897065339617468918
“GPT-4.5 is the first time people have been emailing with such passion asking us to promise to never stop offering a specific model or even replace it with an update great work @kaicathyc @rapha_gl @mia_glaese” / X https://x.com/sama/status/1896231850093551878
“Visited Oak Ridge today with @SecretaryWright @SenatorHagerty @RepChuck, as part of an event where 1,000 scientists are applying OpenAI models to their work. Inspiring to see the ways that people are using OpenAI tools to advance science for America and humanity. https://x.com/gdb/status/1895573890967224622
“Apparently the main thing we’re getting with GPT 4.5 in exchange for a 30x price increase is fuzzy stuff like “EQ”. The ironic thing is this is an aspect of behavior, not capability. My bet is that any differences in EQ are due to post-training, not the parameter count!” / X https://x.com/random_walker/status/1895494391466475684
“Explore full GPT 4.5 results at: https://x.com/lmarena_ai/status/1896590159111713050
“GPT-4.5 rollout to plus users has started will complete within the next few days” / X https://x.com/sama/status/1897348424984617215
“GPT-4.5 is singularly leading on the Style Control leaderboard, showing its strength in both style and substance. https://x.com/lmarena_ai/status/1896590154871210154
OpenAI plans to bring Sora’s video generator to ChatGPT | TechCrunch https://techcrunch.com/2025/02/28/openai-plans-to-bring-soras-video-generator-to-chatgpt/
Why OpenAI isn’t bringing deep research to its API just yet | TechCrunch https://techcrunch.com/2025/02/25/why-openai-isnt-bringing-deep-research-to-its-api-just-yet/
“Hate to say it but I told ya so. Orion is called GPT-4.5 and not GPT-5, not because big labs suck at naming releases, but because it’s nowhere near as impressive as the cost to train it. Also explains why OpenAI sat on Orion for a while. Seems pre-training has indeed hit a wall.” / X https://x.com/bilawalsidhu/status/1895209449448710645
“@ikristoph just imagine GPT-5 and o4 price.” / X https://x.com/Yuchenj_UW/status/1895313053606142283
“Pro tip: Use the OpenAI Playground to compare GPT-4.5 and other models. Watch how “thoughtful” the GPT-4.5 response is. https://x.com/omarsar0/status/1895504181789937964
“Elon: we MUST outpace OpenAI to survive! No slack for charity… but we’ll release our obsolete 2024 model in uh, months after shipping a new one Wenfeng: I don’t really think about competition. Here is our best model. Also have our file system from 2019, it’s still SoTA anyway” / X https://x.com/teortaxesTex/status/1895392169600635146
“so GPT 4.5 is 10x bigger than 4o and only marginally better at most things my read: could be the beginning of the end for scaling laws what happened here? did we run out of data? or do scaling laws just not capture model behavior on tasks we really care about?” / X https://x.com/jxmnop/status/1895525157101584436
“the launch of GPT-4.5 feels a lot like when ChatGPT first came out in 2022 – everyone’s back to having fun just chatting with an AI again” / X https://x.com/stevenheidel/status/1895541898137456776
“@emollick Fair enough. To make my claim more precise, in the examples of supposedly better EQ I’ve seen, I think GPT-4o or even GPT-3.5 is quite capable of showing the same behavior as GPT-4.5 if post-trained appropriately. https://x.com/random_walker/status/1895499480902013254
“Rollout complete ✅ (Faster than expected)” / X https://x.com/OpenAI/status/1897362939340083385
“GPT-4.5 is ready! good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i’ve sat back in my chair and been astonished at getting actually good advice from an AI. bad news: it is a giant, expensive model. we” / X https://x.com/sama/status/1895203654103351462
“You could also call this model “GPT-4o chonky” I think it’s literally just GPT-4o x 10 which would put us again into the ~5T param range” / X https://x.com/scaling01/status/1895415262486388906
“I think the perception of the response is subjective but I feel like GPT-4.5 often has that characteristic of sounding more “thoughtful” — in this case, simply by adding sensations, thoughts, etc.” / X https://x.com/omarsar0/status/1895504558669127693
“METR received access to an earlier checkpoint of OpenAI’s GPT-4.5, 7 days before release. We ran quick experiments to measure the model’s performance. As with OAI’s results, GPT-4.5 performs above GPT 4o but below o1 or Claude 3.5 Sonnet, with a time horizon score of ~30 minutes. https://x.com/METR_Evals/status/1895381625585967180
“Claude 3.7 beats GPT 4.5 most tasks But 4.5 has better vibes… the first non-Anthropic model since 3 Opus 4.5 is vibey, and legitimately the first time a model made me laugh Humor is intelligence You exist in the context of all in which you live and what came before you meaning https://x.com/dylan522p/status/1895557873712972138
“Been using GPT-4.5 for a few days and it is a very odd and interesting model. It can write beautifully, is very creative, and is occasionally oddly lazy on complex projects. Feels like Claude 3.7 while Claude 3.7 feels like GPT-4.5.” / X https://x.com/emollick/status/1895209046925574631
“Sonnet 3.7 is almost too eager and GPT-4.5 is almost too deferential” / X https://x.com/marktenenholtz/status/1895316983144685978
“Brett Adcock talks about Figure parting ways with OpenAI to focus on training their own AI foundation models for humanoid robots. https://x.com/TheHumanoidHub/status/1896267571474911571
“Vibe coding app with Qwen QwQ-32B in a few lines of code https://x.com/_akhaliq/status/1897774596515938394
“🔥 Just released: QwQ-32B achieves DeepSeek-R1’s performance with just 32B parameters (vs 671B)! This breakthrough in Reinforcement Learning scaling proves smaller models can match giants in reasoning & problem-solving. https://x.com/fdaudens/status/1897365520728605153
“Welcome, Qwen QwQ-32B! 👋Excited to have the latest @Alibaba_Qwen in the Arena ready to chat with everyone. https://x.com/lmarena_ai/status/1897763753417900533
“Qwen 32B QwQ – no. 1 trending on Hugging Face – SoTA after SoTA, the competition is heating up! 🔥 GG @Alibaba_Qwen https://x.com/reach_vb/status/1897974348503208081
“developers can get started with pip install –upgrade “ai-gradio[huggingface]” export HF_TOKEN import gradio as gr import ai_gradio gr.load( name=’huggingface:Qwen/QwQ-32B’, src=ai_gradio.registry, coder=True, provider=”hyperbolic” or fireworks-ai ).launch() github:” / X https://x.com/_akhaliq/status/1897775110033227860
QwQ-32B: Embracing the Power of Reinforcement Learning | Qwen https://qwenlm.github.io/blog/qwq-32b/
“Breaking: OpenAI just released GPT-4.5, the startup’s largest AI model to date. Available now to Pro ($200/mo tier) users and developers on paid tiers via API. Everything else you need to know about the highly-anticipated launch: https://x.com/rowancheung/status/1895202496907546718
“OpenAI o1 and o3-mini are now available in the API for developers on all paid usage tiers. Use them with: 🌊 Streaming ⚙️ Function calling 🗂️ Structured Outputs 🧠 Reasoning effort 🤖 Assistants API 📦 Batch API 👀 And for o1 only, vision. https://x.com/OpenAIDevs/status/1897414494286176333
“I’m honestly wondering what people do with it. Because for coding it’s miles behind 4o and o3-mini. Are you guys just generating greentexts all day?” / X https://x.com/scaling01/status/1897590986278117758
“Just noticed ChatGPT Code Interpreter (the OG, with file downloads; not the in-browser kind) not only works in 4.5, it works in o3-mini too. See example using both. Was this announced? C.I. used to be stuck in 4o which was often too dumb or too slow; now there’s a fix for both https://x.com/goodside/status/1897412604894789692
“Deep Research is doing interesting stuff. The o3 model can’t access images directly so, when I asked it where a photo was taken, it installed & used (?) a bunch of Python tools and Sherlock Holmes-ed an analysis. Right answer on two tries, but the clues it describes don’t exist? https://x.com/emollick/status/1896947701608264123
“actually very surprised with this topping every single category given 4.5 doesn’t have use any test-time compute. pretraining is still alive my friends!” / X https://x.com/willdepue/status/1896603985861378440
“@jeremyphoward what I am curious about: is GPT4.5 more expensive and slower than o1 (assuming o1 is GPT4-sized + inference-compute scaling)? Would be interesting in the context of what GPT4.5 with o1-style inference-compute scaling will look like.” / X https://x.com/rasbt/status/1895496476056559811
“GPT-4.5 release shows LLM pre-training scaling has plateaued: A much bigger model, trained with 10X more compute, can only give us so much (or so little?) improvement. That’s why xAI can catch up within 2 years. DeepSeek drastically reduced GPU requirements by optimizing” / X https://x.com/Yuchenj_UW/status/1895531031475920978
How we think about safety and alignment | OpenAI https://openai.com/safety/how-we-think-about-safety-alignment/
Judge denies Musk’s bid to halt OpenAI’s for-profit shift, fast tracks trial | Reuters https://www.reuters.com/legal/us-court-denies-musk-preliminary-injunction-his-suit-against-openai-2025-03-05/
“I have been impressed by GPT-4.5’s vision ability. It can differentiate and count much better than any other model. It even spotted the butterfly. https://x.com/emollick/status/1895211249656570258
“@bobmcgrewai I think that’s an apple-to-oranges comparison. We don’t now how GPT4.5 looks like with inference-compute scaling. Train- and inference-compute are two orthogonal ways to improve LLMs. https://x.com/rasbt/status/1895504882561597817
“Agile teams can’t plan two sprints in advance yet OAI has somehow planned 2027 for AGI” / X https://x.com/abacaj/status/1897771754845573333
