Forgot your password?
Close
typodupeerror
AI

OpenAI Says Its New GPT-5.5 Model Is More Efficient and Better At Coding (theverge.com) 56

Posted by BeauHD from the new-and-improved dept.
OpenAI released its new GPT-5.5 model today, which the company calls its "smartest and most intuitive to use model yet, and the next step toward a new way of getting work done on a computer." The Verge reports: OpenAI just released GPT-5.4 last month, but says that the new GPT-5.5 "excels" at tasks like writing and debugging code, doing research online, making spreadsheets and documents, and doing that work across different tools. "Instead of carefully managing every step, you can give GPT-5.5 a messy, multi-part task and trust it to plan, use tools, check its work, navigate through ambiguity, and keep going," according to OpenAI. The company also notes that GPT-5.5 will have its "strongest set of safeguards to date" and can use "significantly fewer" tokens to complete tasks in Codex. GPT-5.5 is rolling out on Thursday for Plus, Pro, Business, and Enterprise ChatGPT tiers and Codex, with GPT-5.5 Pro coming to Pro, Business, and Enterprise users.

OpenAI Says Its New GPT-5.5 Model Is More Efficient and Better At Coding More | Reply

OpenAI Says Its New GPT-5.5 Model Is More Efficient and Better At Coding

Comments Filter:

  • Sure (Score:5, Insightful)

    by nospam007 ( 722110 ) * on Thursday April 23, 2026 @05:26PM (#66109350)

    My butcher says, meat is healthier than bread and my baker says just the opposite.

    I eat both with a grain of salt.:-)

    • My butcher says, meat is healthier than bread and my baker says just the opposite.

      I eat both with a grain of salt.:-)

      By both do you mean, meat and bread or your butcher and baker? The latter seems low in bread.

    • Sure, everybody touts their own products. But OpenAI has some reason to brag.

      In my own comparison tests of coding LLMs, I've found Anthropic and OpenAI models superior. And OpenAI's are much faster, with similar results, than Anthropic's.

      It's not *just* hot air.

      • i love it... this is like back in the old days when the whole world moved on to trains, planes and automobiles and the cart rights were arguing over which modern breed of horse was best at transporting large amounts of grain.

        Do me a favor, get or rent a GPU somewhere and run your own instance of Qwen 3.6 35B A3B. But make sure you give it the Playwrite MCP and for a bonus, toss in a web search MCP. I think you'll find that the other two still have a speed advantage as far as tokens spent. But that Qwen fini

        • LOL, if you must have qwen for coding, then the dense 27B 3.5 is noticeably better. For "agentic" task management both work, but you can also get a working setup with a much less demanding and faster local model anyway, provided there are things you need to manage with it.

        • I have been using this very model this week. It walks, not runs, on my 5950x with 64gb ram and 8gb 3060ti gpu. Takes a minimum of 3 mins to respond to any prompt. 10+ if it writes code and files. Results vary greatly. Often takes 10 prompts to fix a small bug. Which means hours. Claude can do it right, in minutes. And then i exceed the limit, and am stuck for the next 6+ hours. Or worse, a full week.

      • "Sure, everybody touts their own products. But OpenAI has some reason to brag."

        EVERYBODY has a reason, usually it's to earn more money.:-)

        • My point was, their product is *actually* good.

          Every business is in business to make money, and nearly all of them advertise their products. So that's not unique to OpenAI.

      • I've found OpenAI and Gemini leapfrogging each other. In some cases, I like using both. I have tried the other models, but I seem to get the best bang for my buck with ChatGPT and Gemini, maybe throwing in other models every so often.

        Often, Gemini is faster in my use cases, but both are good.

  • I am still waiting for someone to announce that the new version of their product is worse than the old version at something.
    • In the Age of Enshittification, it would be newsworthy if something new is truly better than its predecessor, but I'm not going to take the word of the company that makes the product.

      • Re: (Score:1)

        by dfghjk ( 711126 )

        "In the Age of Enshittification..."

        The age of Cory Doctorow grift? Do you get paid extra for the capitalization?

        "it would be newsworthy if something new is truly better than its predecessor, but I'm not going to take the word of the company that makes the product."

        A real historian! You don't even know what the term means.

        • Do you get paid extra for the capitalization?

          Of course, I'm a capitalist, baby! I only do things for money. If it bothers you so much, you can pay me more to switch back to lowercase for future posts.

  • That was before we had an "LLM C Compiler" that was bullshit and didn't have an assembler or linker (Anthropic). That was before we had the Mighty "Mythos" that found three whole bugs (supposedly) and threatened a flood of more of them.... that never came. See, thing is, I get curious when people make these claims and tend to try to repeat any really impressive tasks, like when I tried Anthropic's C compiler that turns out cannot link, optimize, assemble, or do anything really well without som

    • Re: (Score:2)

      by Junta ( 36770 )

      Well the 'nice' thing about this sort of language is it can frequently be true multiple times. It's "better" but how close to "good enough" is unspecified.

      The Anthropic one was interesting because the original person behind it was fairly nuanced and honest. The stunt needed an existing reference implementation as a basis as well as a boat load of unit tests and needed hand holding and still didn't quite pass the big test of compiling the kernel (needed to borrow missing bits that claude couldn't figure ou

    • Re: (Score:1)

      by gweihir ( 88907 )

      Indeed. It is all smoke and mirrors and abysmally bad business numbers. I have decided to amuse me with doing some research into AI failing at things in the meantime.

    • I'm curious how an "AI" C compiler compared to a basic optimizing C compiler like GCC. For some tasks, I don't think it will make much, if any difference. For other tasks where the AI can look at the gestalt of the entire program with all linked modules and "knows" what is going on, perhaps it can find ways to fix things (like using a quicksort when someone had a bubble sort), but overall, I'd not be surprised if there is little to no gain over GCC.

      Same reason why AI isn't cranking out "hand tuned" assemb

  • Translation: Still sucks (Score:3, Funny)

    by gweihir ( 88907 ) on Thursday April 23, 2026 @05:58PM (#66109402)

    Just a teeny bit less. Not that the mindless fans will care.

    Oh, and how are those revenue numbers? Still "certain death soon" level?

    • how are those revenue numbers?

      They seem to care more about investors than subscriber revenue.

      • Re: (Score:2)

        by gweihir ( 88907 )

        But investor money is not "revenue". Investor money is something that gets burned. It can produce revenue as an effect (called "ROI", i.e. Return on Investment), and in a good investment, that revenue will be higher than the money invested. But LLM operations seem to be relatively stable at around 15...20% ROI, which is absolutely abysmal.

        • Re: (Score:3)

          by dfghjk ( 711126 )

          "But LLM operations seem to be relatively stable at around 15...20% ROI, which is absolutely abysmal."

          Depends on where in the investment lifecycle you are. AI is permanently at the very beginning while claiming to be imminently mature.

          • Re: (Score:2)

            by gweihir ( 88907 )

            It is. As any good "constant delivery scam" does, where the next version or surely the one afterwards will make it all worthwhile! We just need some more money in the meantime to make that happen. Oh, and we need more money actually, but the payoff will be fantastic! Obviously, that never happens.

          • What AI brings to the table is better software. The next step is having hardware that can feed the LLM what is needed via sensors and allow the AI to control lots of parameters on the machinery. For example, being able to notch a belt tensioner a bit tightly to compensate for wear and chain stretch, or rate/de-rate loads on gears due to the detected wear on the cogs. If it is a transmission, if it can't shift into a certain gear, create different shift points to work around it until the transmission can

  • more faster (Score:3)

    by awwshit ( 6214476 ) on Thursday April 23, 2026 @06:38PM (#66109448)

    Now with more slop delivered faster!

    • Re:more faster (Score:4, Interesting)

      by Tony Isaac ( 1301187 ) on Thursday April 23, 2026 @11:55PM (#66109650) Homepage

      The latest coding models have moved beyond slop. They actually write decent code.

      Just a few months ago, I used to have to micromanage every code change. These days, with GPT-5.4, it usually gets it right the first time, even larger code updates. It does a great job of following the coding patterns and conventions YOU demonstrate in your code base. It's actually not hard to read or...sloppy.

      • > It does a great job of following the coding patterns and conventions YOU demonstrate in your code base

        I didn't need AI for that, why now? And maybe I want it to write code better than I can.

        • > It does a great job of following the coding patterns and conventions YOU demonstrate in your code base

          I didn't need AI for that, why now? And maybe I want it to write code better than I can.

          Maybe you want to to write code faster than you can? I used Claude to build an entire an entire web-based application with SQL Server backend from scratch in less than a day. Is it the *best* code that could possibly be written? Probably not. Does it work as intended? Yes.

        • In many cases, it actually does write code better than you yourself would. It follows your patterns, but it doesn't forget to do things like handle the possibility of null references, something I myself inappropriately skip on occasion.

          It definitely writes code *faster* than you can. And that by itself is a big benefit.

  • So this one is a bit better than the last "I see now what I did wrong. Let me fix that for you"

    But I thought you told us that by now no programmer would
      have a job ?
    And I still have one.

    Go bust already, Dirty Sam and friends.
    • If you canâ(TM)t see the change happening in real time in this profession, youâ(TM)ve got bigger problems than AI taking your job
      • I saw change, I use AI tools every day, I am not seeing big gains with newer models.

        I believe LLMs are plateauing, more data and more parameters will not lead to notceably better tools.

        AI tools are useful. The AI crash is coming.
    • Two things can be simultaneously true: 1) Many of the AI proponents and CEOs exaggerate and overhype the power and usefulness of their software. 2) These systems are powerful and continue to get more powerful.
      • Indeed. At the moment, I am simultaneously amazed and appalled by the answers given by AI. It's like having a conversation with someone with Multiple Personality Disorder where some personalities have an IQ of 120 and other personalities have an IQ of 80, and the person is really good at disguising which personality I'm currently engaging.
      • I use AI every day. But despite repeatedly being told how wonderful the latest models now are, and seeing stories of people ( usually with their own AI tools to sell ) who claim not to write code any more, even moderately complex problems seem beyond the AI. While, worse, the AIs keep claiming to understand problems which they clearly do not.

        "These systems are powerful and continue to get more powerful."
        We are not seeing this. I think LLM usefulness for coding has plateaued and feeding them more data and ad
  • AI models produce slop code, but that's what programmers have been producing for years anyway, now it's just ever so much moreso. When autocomplete was invented, it was great, but then sloppy programmers wrote code that couldn't be understood without autocomplete tooling.

    Now we have 99% of the people in the industry who joined for the money, not because they enjoy programming. They write more slop.

    AI is just the next iteration. Doesn't work in edge cases? No problem, write a ticket I'll fix that. I'm ju
  • This would be great except, just as the geniuses of AI predicted, I and all other programmers lost their jobs to AI a year ago.

    Oh, no wait.
    That was utter bullshit.

    I still have a job and GPT 5.n+1 is 0.1% better than GPT 5.n, because LLM improvements have plateaued.

    Dirty Sam is fucked.
  • proclaims that they're latest version is better than the last one.

    That's the entire point of a new version.

    When was the last time a vendor released a new version and they're marketing team was all "Hey fam, guess what? New version is out, and it sucks sweet ass!"

    Why does the crack /. editorial team need this explained to them?

  • Does it come with murderous intent built in or doesn't it understand the effect it has when controlling autonomous weapons systems?

Slashdot Top Deals

BASIC is to computer programming as QWERTY is to typing. -- Seymour Papert

Close