Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

MAI-Code-1-Flash

microsoft.ai

Post reply on HN
Print view View original

151–160 of 297 posts

Re: MAI-Code-1-Flash

#151

Post by hashmap »

ggcr wrote: post #91
giancarlostoro wrote: post #69

Mark Zuckerberg must be in crisis. Microsoft releasing models that compete with Claude's models. Meanwhile the only thing anyone knows about Mark's models is that they help you get hacked more easily.

Meta recently launched Muse Spark [1] and they themselves compare against Claude Opus 4.6 Max. Here Microsoft is comparing against Claude Haiku, the smallest and least capable model from Anthropic. [1] https://ai.meta.com/blog/introducing-muse-spark-msl/

i have had good results adding muse spark's contemplate mode as a roundtabler for complex questions. but you cant turn off their data ingestion for training so that is a shame.

Re: MAI-Code-1-Flash

#152

Post by girvo »

Flere-Imsaho wrote: post #39
redrove wrote: post #27

Earlier quoted context omitted.

It’s about bang for buck. That high a score for 5B params is pretty good, nigh unbelievable a short while ago. It is my belief that smaller models will get better and better, and even cloud SOTA models will shrink. Yet another reason the current buildout will feel like the railroads.

Yeah the future is probably a number of highly specialised small models you can run on your own hardware rather than massive frontier models in the cloud. That's what I'm betting on anyway.

Step 3.7 Flash on my Asus GB10 based mini pc is incredibly close to that today. I’m very impressed, and that’s without MTP to boost performance

Re: MAI-Code-1-Flash

#154

Post by fnordpiglet »

bel8 wrote: post #82

It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub Copilot which I cancelled yesterday. GitHub Copilot had competitive pricing until yesterday when they changed from per-request to one of the most expensive per-token quotas. Seriously, take a…

I use larger models to organize work into a topologically sorted task graph and pin smaller models to the tasks depending on the complexity with a larger model evaluating the work and patching where necessary. This uses haiku quite often for routine work. I’m able to do multi hour highly complex work with superior results and a much lower bill as a result by doing this, with a parent orchestrator able to do a massive labor within a single context window by effectively organizing work and reviewing quality and integrating where needed. I don’t use haiku directly, but it’s often 30-40% of any major efforts token use. This further improves time to completion as well as cost - but I find haiku is better at following literal instructions and plans without “second guessing,” while opus class models second guess in their thinking constantly.

As such, haiku isn’t a waste of my time, it saves enormous amounts of time for me. But I spent a large amount of time building the orchestration system up front and iterating on it to get here. Interestingly i found my experience as a director and later a distinguished engineer gave me the tools to build it and get it working well and reliably end to end - the dynamics of multi agent workflows of varying capability is not a lot different than the dynamics of a 1000 engineer organization.

Re: MAI-Code-1-Flash

#155

Post by HDBaseT »

whalesalad wrote: post #104
efields wrote: post #48

Please test your websites in Safari. Almost all of your iOS users use it by default, and the desktop experience is pretty close to the mobile experience, so testing is easy. That scroll effect is jank city for me (yeah yeah works fine in Chrome/Edge).

some kind of scroll hijack going on for sure, feels terrible on firefox+macos

I instantly close websites which use this weird scroll hijacking and slow animation nonsense.

Let me slide as fast and unrestricted as I want. I do not want to "transition" to the next paragraph.

This trend needs to stop.

Re: MAI-Code-1-Flash

#157

Post by lukevp »

fnordpiglet wrote: post #154
bel8 wrote: post #82

It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub Copilot which I cancelled yesterday. GitHub Copilot had competitive pricing until yesterday when they changed from per-request to one of the most expensive per-token quotas. Seriously, take a…

I use larger models to organize work into a topologically sorted task graph and pin smaller models to the tasks depending on the complexity with a larger model evaluating the work and patching where necessary. This uses haiku quite often for routine work. I’m able to do multi hour highly complex work with superior results and a much lower bill as a result by doing this, with a parent orchestrator able to do a massive…

Got anything from your orchestrator you could share that’s usable by others? Sounds like how I’d like to work but is difficult to get going from scratch

Re: MAI-Code-1-Flash

#158

Post by SwellJoe »

bel8 wrote: post #82

It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub Copilot which I cancelled yesterday. GitHub Copilot had competitive pricing until yesterday when they changed from per-request to one of the most expensive per-token quotas. Seriously, take a…

I've been doing benchmarking of various models for finding hard security bugs, and my faith in Haiku (and Sonnet, even) has dropped precipitously in the process. Self-hosted Qwen 3.6 27B consistently outperforms both for finding security bugs, which was a shocking result. I expected Qwen to be around Haiku level, maybe a little worse, and I definitely expected it to be worse than Sonnet.

And, DeepSeek and MiMo perform much better than Haiku and Sonnet, near Opus/GPT 5.5 levels, at a fraction of the cost.

There's seemingly no reason to ever use Haiku or Sonnet, if you're not getting it for free or as part of a subscription (that you don't usually saturate).

Re: MAI-Code-1-Flash

#159

Post by perching_aix »

tedggh wrote: post #126
hparadiz wrote: post #106

Earlier quoted context omitted.

The $20/month ChatGPT plan that comes with codex is good value. Even just have premium ChatGPT is nice. I get rate limited regularly but it still lets me do most things.

The $100/month is excellent value. I don’t understand how’s that not the default option for all professional developers. Unless people don’t produce any value writing code, like playing around and experimenting with vibe coding, I understand. But if software development is your actual income, and assuming you live in a wealthy country, $100/month is nothing for a tool like Codex.

Picked up the most recent SO developer survey that features relevant info, the 2024 release: https://survey.stackoverflow.co/2024/work#coding-outside-of-...

The supermajority of respondents did report that they do engage in some coding outside of working hours, for one reason or another. I'm impressed; I'm basically a zombie after hours, rarely in any shape to touch anything technical. Good for them.

But then only 19.3% of respondents ticked that they code for freelancing reasons, and only 15% said they're doing it in an attempt to bootstrap a business. These groups were the only types that suggested revenue generating after-hours activity, and they even overlap to a non-obvious-to-me extent. But even if we pretended they didn't, that adds up to like a third at best.

So when you say:

> I don’t understand how’s that not the default option for all professional developers.

that's in contradiction with this data (and imo common sense), which suggests that the supermajority of professional developers simply do not perform revenue generating software development activity outside of work hours, period. Therefore, for them, the ROI on any potential AI subscription is a flat and constant zero.

Unless you envision people working at "bring your own license" type shops, I don't know how this is supposed to make sense. These are work tools, corporate should be providing them already. But then I'm clearly not from a "wealthy" country either, so YMMV.

Re: MAI-Code-1-Flash

#160

Post by flumpcakes »

GaryBluto wrote: post #99

What's with the lack of Microsoft design language on the website? It's painfully obvious they're trying to emulate Anthropic's style here and it looks tacky.

Thank you! This website is dreadful for accessibility and usability.
Post reply on HN