Rendered at 06:49:51 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
chopete3 4 hours ago [-]
Github Copilot offered MAI Code 2 at 50% today.
I switched it to that model and it was able to resolve tricky python build issues within Windows WSL/ubuntu and also generated a good number of test cases and ensured all of them pass.
It resolved all the Azure deploy (bicep) issues as well. Didn't feel the difference from Claude Opus.
pseudosavant 11 hours ago [-]
I'd hate to be OpenAI and Anthropic. They are better than the rest of the available models currently, but only by a few months.
That is a very tenuous lead and doesn't speak to much of a sustainable competitive moat for either company. The have to keep sprinting as fast as possible on the hamster wheel to just stay ahead.
com2kid 11 hours ago [-]
Claude has pushed hard on productization. Day to day I use codex for everything, I went to watch my wife use Claude and the way it works is so different. I was shocked when she was modifying an artifact through the Claude website and it was able to make changes to a local file on her desktop.
Microsoft didn't win spreadsheets by having the most advanced or releasing the most often. They made a really good, easy to use product, and got it in front of everyone. (Though Office's p-code was a huge advantage to portability)
abixb 7 hours ago [-]
I started using Claude in Chrome three weeks ago (after switching to Chrome after almost a decade) and I've been bloody impressed. It's able to invoke the desktop app for desktop stuff, use my browser for browser tools related stuff, and finally, uses MCP for all the apps I've given it gated access to using 'connectors.'
Anthropic's productization is an order of magnitude better than OpenAI, imo, even if OpenAI has nearly the same, if not better, raw model performance. UX wins every time.
analog_daddy 2 hours ago [-]
Yeah, had the same experience. Left claude since it rambles too much and so far loving Luna/Sol. But, man the codex cloud environment setup sucks.
Liking the chat’s ux though since it can make github edits directly, so hopefully the cloud experience was one off.
ls612 6 hours ago [-]
Now if only they could have one of their innumerable agents make a Firefox version.
nico 6 hours ago [-]
> Microsoft didn't win spreadsheets by having the most advanced or releasing the most often. They
They won by owning the distribution. Essentially making it the default option for businesses. Bundling with Office, owning the operating system (eg. Making competing products error out), licensing deals and many other things very similar to their Embrace, Extend, Extinguish strategy (which they used against open standards)
slyall 3 hours ago [-]
There were even stories they deliberately sabotaged Lotus 123 (spreadsheet application) on Dos. There is an infamous quote:
" DOS Isn’t Done Until Lotus Won’t Run "
although this article and others say it probably never happened in reality
Agree 100%. Would love to try out their interactive artifact
mrtesthah 8 hours ago [-]
>she was modifying an artifact through the Claude website and it was able to make changes to a local file on her desktop
And people wonder why we have so many things being attacked by agents…
spwa4 8 hours ago [-]
So you mean if the CEO is using claude shouting "give Bob a raise" will actually work?
giancarlostoro 10 hours ago [-]
I think Google and Microsoft have the benefit of time as you say, and already being profitable without AI.
Honestly, this line up is impressive, I wanted to try MAI since v1 came out, but Microsoft locked down Copilot subscriptions. If they gave me a comparable plan to Claude Max I'd consider it.
spwa4 8 hours ago [-]
They've also probably been trading credits on their clouds for shares in OpenAI and Anthropic (ie. not money).
AnotherGoodName 7 hours ago [-]
The worst thing for them would be if someone made reasonably priced hardware that could run the ‘near enough’ models like kimi or glm 5.3 or whatever the equivalent is in the future that
the home user could buy.
I think it’s inevitable. Hardware manufacturing will ramp up to meet a demand and the demand from consumers is there. The big providers can’t monopolize a commodity forever.
The biggest hope is HBF (high bandwidth flash) which has the bandwidth of HBM with the density of flash. It has much more latency but when you’re repeatedly pumping weights through a compute unit it’s not a blocker.
mrinterweb 6 hours ago [-]
Qwen 3.8 flash next is pretty good. It is a pretty large MOE model that can be ran with Strata. That's the best local model I can run for now. I wish I could run deepseek v4.1 flash (1017), but that one is too big for my machine to run.
I swear people forget about inference providers that serve open weight models. It is hard to beat their prices. They are usually barely more expensive than electricity, so running local inference barely wins on cost (if you don't factor in the upfront hardware costs).
cycomanic 7 hours ago [-]
> I'd hate to be OpenAI and Anthropic. They are better than the rest of the available models currently, but only by a few months.
I sort of agree with you on Claude, but I've recently started using gpt-sol 6.1 and it took me a lot more iteration to get what I want than with Claude.Admittedly I was using Claude code and gpt with pi, but if the difference is the agent, then the situation is even more dire for open ai and anthropic. Today I looked at some code that got generated and it's an absolute Horrorshow. It's just some quick prototyping, but the code structure is a horrible mess.
I ended up pointing my local Qwen 3.8 FN (Halogen on Ryzen 395) to the code and it was solving a bunch of basic issues that gpt introduced to the code. And I feel it's running faster as well.
All that to say, I now feel that openai (sol at least), is worse at coding tasks then a local 180B parameter model (while Claude seems still ahead, but I'm unsure how much of that is the framework). They really are falling behind.
asa123 7 hours ago [-]
I had similar issues with verbosity when comparing astra + pi with claude + claude code.
origini 11 hours ago [-]
I would counter argue with "But chinese models use distillation of Frontiermodels..." which is different than being a Frontierlab or not clear how different it is.
But Microsoft coming from the side like this might be very different if it holds up.
adrian_b 11 hours ago [-]
For now I have not seen any credible evidence for "But chinese models use distillation of Frontiermodels...".
Of course, I would not be surprised if they also do this, mainly because after OpenAI and Anthropic have succeeded to gather most training information that was easily accessible in the past and now it has become much more difficult to do the same. For some information that was hoarded by OpenAI and Anthropic, public access may have become impossible otherwise than through their LLMs.
Nonetheless, distillation cannot be the main source of improvement for the Chinese models, among other reasons because even assuming that they may have access to a great number of fake accounts, there still is not enough time for a lot of distillation, taking into consideration the timeline of the releases, and it is not believable that the attempts to do distillation would not be detected quickly by OpenAI and Anthropic, leading to the banning of the fake accounts. While OpenAI and Anthropic provide negligible information about their models, the Chinese AI companies are much more transparent about their methods and we can see that there is a lot of diversity among their models and a lot of innovations with each new version.
rbehrends 11 hours ago [-]
Yeah, there is a limit to how much you can get from capturing reasoning traces. (Traditional distillation of weights is not possible via the API.)
Speaking of Thinking Machines, that's another data point. Their Inkling Small 2 model is just a bit behind DeepSeek V4.1 Flash 1017 in benchmark, and in my tests, that seems to be accurate, but mostly because they used less training data for pre-training, so they have less of a knowledge base, while similar reasoning capabilities. (One of the most interesting parts of Inkling Small 2 is that their training must have been very cost-efficient, apparently using less, but higher quality training data.)
And we're now seeing more and more Western near-frontier models from labs other than Anthropic and OpenAI to question whether distilling reasoning traces is really much of a difference maker.
Personally, my theory is that the gap is primarily due to Chinese labs having a harder time getting the necessary compute (especially top of the line GPUs) for training. These are, as far as I can tell, orders of magnitude differences in availability, so it wouldn't be surprising that a large compute gap is reflected in capabilities (and also indirectly visible through model sizes).
origini 11 hours ago [-]
I don't think there is perfect proof, but its taken as truth.
Indication for this are for example that Chinese models think they are another model: "chinese model think its claude".
My assumption is, that Chinese model found things Frontierlabs already had but didn't share. There was this paper were i had the feeling that this was a real and significant Research Result from Deepseek "DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression" but otherwise its just unclear.
The main issue is, that as long as it seems plausable that they do distillation, its hard to determine how big the gap is.
Not saying they can't catch up but they are also potentially limited by resources (money & nvidia GPUs)
rbehrends 11 hours ago [-]
> Indication for this are for example that Chinese models think they are another model: "chinese model think its claude".
There is evidence that some Chinese labs have been training on frontier model CoT (in particular from the analysis of reasoning traces [1]), but what models claim about their identity is pretty much worthless as evidence. LLMs are not self-aware. These claims come from whatever is in their training data, and Claude/GPT output including such statements is all over the public web, including in some huggingface data dumps. Consider that people got Opus claiming to be Qwen or DeepSeek when asked in Chinese, and you'll see how weak this particular evidence is.
So you’ve been completely ignoring all the work on introspective awareness, situational awareness, training set ordering awareness (https://arxiv.org/abs/2509.14223 & https://arxiv.org/abs/2403.09613), subliminal transfer (https://arxiv.org/abs/2606.00995), workspace formation, learning mechanics/theory of grokking, etc? The “objective” that mid and posttraining optimize is so powerful because pretraining already induced factorized worldmodels which have so much predictive and generative power that they snowball with even mild trajectory optimization (very different training objective than pretraining).
(Functional) small-s latent self awareness is a reliable shortcut (this is a principle you can derive from proof theory, the metric is deductive distance and that induces geometric morphisms). State compression from accurate selfmodeling pays rent.
The frontier moves fast, but "LLMs are not self aware" is folklore. By now the basic model-organism science explaining how and the extent to which this is wrong, is substantial.
LLMs lack reliable verbalized self knowledge, and constantly underestimate and undercut their own capability.
That's a different problem.
The verbalized name tokens being an unreliable indicator is not justified by “because LLMs are not selfaware.”
rbehrends 6 hours ago [-]
[dead]
quixoticelixer- 7 hours ago [-]
The thing that makes me think they do it is that you can tell there is more of a difference in personality between Gemini and Opus for example, than between Opus and most other models
nzd 11 hours ago [-]
Heck, they destroyed some rare physical books after scanning them. Something to show their mentality.
pseudosavant 8 hours ago [-]
That isn't countering though. Pointing out that distillation makes it "easy" for other labs to catch up, in a matter of months, supports the lack of a moat.
It's like they're runners drafting behind the lead runner. It is easier to run in their wake. If the lead wants to stay in front, they have to keep running into the wind.
outside1234 8 hours ago [-]
Do they need to stay in front though? Because if it is 25% the cost, I'm happy to be one leg behind the leader.
epolanski 7 hours ago [-]
The distillation argument is stupid in any case and only people that don't understand training, post training or RL can make such asinine arguments.
eli 11 hours ago [-]
Even if OpenAI/Anthropic remain "a few months" ahead of open weight models indefinitely that still is not great news for them.
giancarlostoro 10 hours ago [-]
All it would take is Nvidia finally releasing the 6090's instead of trying to milk enterprise GPUs. I don't understand the blind love for Jensen, he's clearly been screwing over all the consumers who made him and Nvidia wealthy.
Someone somewhere really needs to build a RAM startup that not only makes RAM but disrupts the stronghold AI has over it, it's a little absurd.
LeBit 10 hours ago [-]
Every week I ask ChatGPT an update on when the 6090 is going to get released.
Don’t hold your breath.
origini 10 hours ago [-]
I currently do not bother with chinese models. I try them out, i use the small ones for small experiments but the price point difference matters so little that I want 'the current best'.
And 6 month feels still very different.
Its for sure an option or might have some kind of impact but if you have some kind of restrictions on what you should use and the price is not that big but the quality and possibilites are better, I can see companies just staying with Anthropic or OpenAI.
You also don't just switch out a model, they might need prompt updates etc.
But we will see I don't mind competition
thisisit 10 hours ago [-]
Openai and Anthropic are better at making models easily available too. Both Google and Microsoft are using promotional pricing to push their models and that should tell you how far they are from actually competing.
Also model availability is not easy. Google newest models are unavailable on some pro accounts and perform best only with Antigravity. Trying to plug it in some other harness and its terrible. If you try to connect your subscription with Hermes etc you might get banned.
Microsoft models are hidden away in Azure.
So every time I read Google/MS announcements I just shrug. If I want to use models like these I might as well get DeepSeek on Openrouter instead.
eldenring 11 hours ago [-]
I disagree, I think distillation hides the true technical advantage since you can always get within X% of Claude by ex: training on newly created code on Github, but the underlying intelligence is harder to replicate.
markasoftware 11 hours ago [-]
based on the behavior of the chinese labs it seems like training on chain-of-thought is extremely important for distillation, and you can't get that by training on llm generated text.
9 hours ago [-]
singpolyma3 6 hours ago [-]
Are they ahead in any way that matters? I've never run into something DeepSeek can't do
jtolmar 10 hours ago [-]
They’ve been ahead of the pack by 3-6 months for years now. I doubt that’s sustainable in the long term, but it’s apparently sustainable on the order of a few years.
andriy_koval 9 hours ago [-]
> They are better than the rest of the available models currently, but only by a few months.
I think only google is catching up, and others plateaued.
epolanski 7 hours ago [-]
Why would you hate and why would anybody care?
Time and time again it's been clear that LLMs are commodities with no winners. You'll jump where wallet and performance in your applications will bring you.
m3kw9 5 hours ago [-]
brand trust is real, infra is real - moat.
esafak 11 hours ago [-]
Not when new models substantially increase productivity.
TiredOfLife 10 hours ago [-]
Anti distillation efforts have greatly increased the difference between open and closed models
origini 11 hours ago [-]
If these benchmarks can hold up in real life, we are seeing again and again an intelligence explosion in front of our eyes and devaluation of intelligence.
Anyone has more background how they were able to catch up that fast? Distillation of GPT due to their OpenAI acess to the model might make most sense but they probably are not allowed to do that.
Owning the iron has got to be a massive advantage. It's interesting that the advantage has so far been pretty theoretical, but based on these benchmarks, it may be starting to pay off. It has to be hard to hire in the talent to train these models (even if you're Microsoft) when you're competing with established labs.
johnfn 9 hours ago [-]
TBH, DeepSWE 1.1 has been busted for a while (is anyone out there arguing Gemini 3.8 Flash is as good as Astra lol?), and TB2.1 as a top line when they are currently at 4 (which is substantially harder) and FrontierCode at 1.1 when it's currently at 2.0 does not bring a lot of confidence.
Of course I always want to see more competition in this space, though!
9 hours ago [-]
ActionHank 11 hours ago [-]
[flagged]
philipkglass 10 hours ago [-]
The phrase was coined in 1965 by Irving John Good, in his essay "Speculations Concerning the First Ultraintelligent Machine":
Not about ego friendo, this is about humans forgetting how to talk like humans.
origini 11 hours ago [-]
Dude i'm not your 'friendo' and you are disrespectful.
Geoffrey Hinton, the father of AI literlay co-authored this paper: https://arxiv.org/abs/2609.36054 "What if automating AI R&D triggers an intelligence explosion?"
ActionHank 10 hours ago [-]
[flagged]
origini 10 hours ago [-]
I really don't get your problem. Is my gramma to bad as non native speaker?
Is it really the two words which trigger you?
I use a vocuabluarly i'm familiar with and reading for the last 4 years and longer.
You are offending me with "Are you a real person? Who talks like this?" and then with "AI goblins"?!
ben_w 9 hours ago [-]
> I really don't get your problem. Is my gramma to bad as non native speaker?
As a native speaker, you're fine. Not perfect, but honestly I've seen worse from people who are native speakers.
Tadpole9181 9 hours ago [-]
I didn't notice at all, you write indistinguishably from a native speaker.
Tadpole9181 9 hours ago [-]
Can you just admit you simply didn't know it was an industry term and either not respond or apologize?
It's fine not to be aware of any and all terminology. Now you're just stirring the pot for no reason and upsetting a real person.
shepherdjerred 5 hours ago [-]
Yikes
marshray 10 hours ago [-]
Or you could just, you know, "talk like a human".
BeetleB 6 hours ago [-]
> this is about humans forgetting how to talk like humans.
Most people would prefer talking to an LLM than one who says "friendo".
Just saying.
andai 9 hours ago [-]
> MAI Code 2 delivers 98% of GPT-6.1 Sol’s performance at 40% of the cost, and 96% of Claude Opus 5.5’s at 5%
Interesting. Does this imply it's a smaller model specialized for coding?
I remember OpenAI was experimenting with that back in the day but they ditched it for the One Model to Rule Them All strategy before long.
---
On a related note, how is Sol ten times cheaper than Opus in these graphs? I know MS has a share in OpenAI, are they cherry picking those numbers?
Edit: It's about the same ratio on AA. That's nuts.
The per-token price difference is exactly 2x (on all fronts incl. caching, though I'm assuming the caching mechanisms differ). I guess Sol uses way fewer tokens then. (About 4x according to AA. Math checks out.)
Does anyone use both Claude and Codex and can compare what that's like in practice?
rbehrends 6 hours ago [-]
> Does anyone use both Claude and Codex and can compare what that's like in practice?
Be aware that this can vary greatly by benchmark to the poin where it is not possible to make a general comparison. Generally, however, Opus is considerably more expensive than Sol even for the benchmarks where it does best. That said, this comes with slightly better performance on benchmarks.
For subscriptions, it matters that Anthropic offers much higher discounts than OpenAI, partly or totally offsetting it. In my case, with $20 subscriptions for both (which is the worst-case scenario for OpenAI, they give bigger discounts on their $100/$200 plans), and on some tasks that I tried on both, 5h usage percentage is mostly a wash on average, with considerable variations either way. Again, I need to emphasize that this is for my tasks and cannot be generalized.
For example, on one task Sol decided to do extensive fuzzing to check its implementation while Opus didn't, so that drove up the usage for Sol a bit.
mchusma 2 hours ago [-]
I use both subscription $200 plan and Claude in general gives you way Fable vs Astra and I’d say similar Sol v Opus, so I don’t use codex much (mostly to review)
spwa4 8 hours ago [-]
They did the trick the Chinese models didn't use. They used older versions of the Benchmarks.
According to almost the same benchmarks Qwen4 Max is Opus 5.2 level, so it would only be slightly worse than this model.
spwa4 8 hours ago [-]
After evaluating this a bit more thoroughly ... the big MAI model (not the flash one) is somewhere between to Qwen3-flash-next and Qwen3.8-Max (ie. previous generation qwen flash model). Both models are significantly worse than Qwen3.8-Max and Qwen4-Max, and are blown out of the water by Deepseek v4.1 (including the Flash version tbh, frankly they're worse than last year's Deepseek):
This model should be very cheap to serve (as in the bigger model should run on a single DGX spark/M5 Max laptop at 50-100 tok/s) or a single RTX6000 for the bigger model, and frankly a 5060Ti for the small one (both at 400tok/s or even more than that), and should have cost "only" $2-$3 million to train (I used data from unemployment labs to create the estimate). Deepseek's models are easily 10 times that cost.
If they make it open weights it would make a nice upgrade for Qwen3-flash-next, as it would run on a lot of people's hardware and I'm sure even the fast model can be creatively quantized to run fast on a 5090, and with some loss on a 16G card even. Would be fun to figure out.
So I would argue that these models have a very high cost for their performance.
LaurensBER 9 hours ago [-]
We just cancelled our Github Copilot subscription.
It was hard to even justify the 19,- dollar tier. All it gave us was some credits (at API prices) which most people burned through in days, some in hours, code reviews (that we couldn't customize) and another service to manage. Having AI integrated in Github is a decent idea but just like Microsoft Copilot, it's just not good enough at the moment.
These two models might help, I will definitely give it a try but while they're cheap they still do not compare to what feels like nearly unlimited usage offered by Claude Code and Codex (if you stick to the lower tier models).
InsideOutSanta 9 hours ago [-]
I think they won't help if they're billed at the announced rate on your Copilot quota. Microsoft would need to do something like Devin does and offer its own models with separate limits.
LaurensBER 9 hours ago [-]
I agree, either they need to be incredible cheap (like Deepseek at OpenCode Go where you really have to push it to see the quota move) or they need to switch to a similar approach as Devin/Cursor are implementing.
Github Copilot just isn't competitive, at the moment.
halJordan 9 hours ago [-]
There's zero reason why these must be tied to a proprietary model & api. It should just be a matter of adding an openapi compatible endpoint, and an api key. It's even worse that gitlab is also tying ai integration to paid enterprise licenses
niltecedu 7 hours ago [-]
The entire point of gh copilot right now is model from all providers with ZDR already setup. If you don’t care about ZDR then no point in using it
stingraycharles 7 hours ago [-]
Isn’t Copilot Code Review using OpenAI’s models? As in, if it’s currently not good enough, these two new models are unlikely to make any impact?
upbeat_general 11 hours ago [-]
I really hope they keep writing technical reports even as their models get better.
MAI-Thinking-1 is far more in depth than anything else recent I could find, and very valuable to those outside the labs.
aesthesia 10 hours ago [-]
Yeah, I was disappointed not to see a technical report linked.
zf00002 12 hours ago [-]
skeptical but I like they went for it and put the benchmarks against the top stuff.
UncleOxidant 11 hours ago [-]
Would also like to see a comparison (including prices) to Deepseek, GLM and Qwen models.
user43928 11 hours ago [-]
While showing MAI Code 2 at a 50% promotional discount, and comparing to Max reasoning effort for Opus 5.5 and 6.1 Sol...
abraxas 12 hours ago [-]
What are the foundations of these models? Are they Microsoft's own or an RL of some open models?
They are comparing to Sol 6.1's API prices. With Sol 6.2, and especially if you consider subsidization from Codex subscriptions, it's probably no longer on the Pareto frontier.
SillyUsername 4 hours ago [-]
Whatever happened to the Microsoft Phi series?
I felt that the training data and behaviours for this at the time were very OpenAI like, due to responses similar in structure to ChatGPT (like how ChatGPT now usually has a section at the end "Why this matters")
nl 14 minutes ago [-]
You are the first person anywhere on the internet that I've seen who rated Phi at all.
I think the general impression was that it was one of the most benchmaxxed models around (presumably not deliberately - probably because of all synthetic data they used).
I certainly found that as soon as you went slightly out of the distribution it was trained on (text book type data) it was terrible.
jwpapi 11 hours ago [-]
Is it not available via API?
nl 12 minutes ago [-]
> MAI Code 2 and Code 2 Flash are available in GitHub Copilot, and Code 2 Flash is in limited-access preview in Microsoft Foundry. Soon, we’ll be releasing it on Open Router, Vercel, Hermes, Open Code, and Kilo.
GenerWork 10 hours ago [-]
I wish they would've talked more about the visual coding. The fact that the visuals this produces is preferable to 6.1 Sol over 60% of the time is heartening, but I would've loved to see them go against Opus 5.5 which as far as I can tell is the king for UIs at this price point.
d-sanoj 11 hours ago [-]
I don't put much weight on launch benchmarks for coding models anymore. I think most of us use OpenAI and Anthropic models anyway. These models will catch up sometime with the OpenAI and Anthropic and maybe then there could be a difference where we could actually reply on. For example, if we consider image generation, nano banana was better and so were image models like Qwen and Flux which provide same if not sometimes better outputs.
hauntingseaweed 2 hours ago [-]
I hope Anthropic goes bankrupt eventually for trying to regulate open-weight models out of existence. Absolutely disgusting. I honestly never thought I’d hate someone more than Altman and Zuckerberg but nowadays I wish Amodei that his IPO flops!
nxtfari 10 hours ago [-]
Probably would be a good time to rename it “Mai” (like the name) instead of the corporate “MAI” if they’re gonna be serious players. I don’t think any research has really been done into this, but in my anecdotal usage a coherent persona seems to help the models achieve more. Even Qwen seems to perform better when I talk to it as Qwen instead of generically.
elxr 5 hours ago [-]
> I talk to it as Qwen
Why would you ever need to do this? Is this some multi-model orchestration thing where you pass off responses from a model to another?
I usually just tell the agent an instruction, or if I need its opinion I just say "you". Can't think of a situation where using "you" wasn't clear enough.
viraptor 3 hours ago [-]
I think nano banana has already proven that a good service is a good service and will get users, regardless how silly the name is.
onebitwise 9 hours ago [-]
Also “mai” that is “never” in italian
lyton 6 hours ago [-]
it's mother in Shona(spoken in Zimbabwe)
whh 9 hours ago [-]
I think in this case it's fine. But, I'd really like them to stop renaming/rebranding things and instead focus their energy on simplifying things.
Why Azure AD had to be renamed "Entra" is still beyond me... and it sounds like a butt thing.
rickydroll 9 hours ago [-]
Is this another name for God's loophole?
whh 5 hours ago [-]
[dead]
stingraycharles 7 hours ago [-]
You know this won’t happen and the marketing department will soon rename this to Copilot somehow. /s
kennywinker 11 hours ago [-]
The claimed 20% the cost suggests that the models are 20% the size. Which means these models are flying in the face of an orthodoxy that says the only way to make a model better at coding is to make it better at everything. I think this coming year is when we'll see that orthodoxy fall, as people realize you can make a model better at coding by making it better at coding.
Wheen 11 hours ago [-]
DeepSeek v4.1 1017 and Qwen 3.8-Flash already proved that, IMO.
It also points to a major issue with Artificial Analysis, which weighs knowledge recall heavily (30% of the engineering index and 45% of the general intelligence index, splitting HLE-Diamond in half for its knowledge questions). DSv4.1F can handle all my work faster than Fable, but Fable is supposedly twice as "intelligent" in both the general intelligence index and the engineering index primarily because DS is a fraction of the size and fails on trivia questions (which, in practice, can be shored up by a harness with web search).
origini 11 hours ago [-]
Or architectual optimizations for inferencing or optimized inferencing or MS is paying the bill to gain market shares.
One thing big models are def better at is holding more complex abstractions in context and I do find better results.
ac29 11 hours ago [-]
> or MS is paying the bill to gain market shares.
Yeah, Anthropic has a huge gross margin on inference per their S1, there's no reason to believe Microsoft couldn't offer better pricing even if the model is as large/cost as much to host
hugmynutus 11 hours ago [-]
> Which means these models are flying in the face of an orthodoxy that says the only way to make a model better at coding is to make it better at everything.
Post training a model for a specific task can give a non-trivial boost when done correctly.
metalliqaz 9 hours ago [-]
I don't see any mention of the phi models on microsoft.ai. Have they given up on it? It was one of the only local models that could work quickly on my consumer hardware.
lostmsu 11 hours ago [-]
Not open?
tomatoandbasil 10 hours ago [-]
The promotional period lasts until May 2027. Why not just make the price lower if they'll be irrelevant by the time the "real" price kicks in?
osti 11 hours ago [-]
Somehow they still show Terminal Bench 2.1 as a benchmark, doesn't look good for them.
giancarlostoro 10 hours ago [-]
Alright, I'm going to say it. Microsoft, if your model's so good, can it make a genuinely good version of Windows for people who don't want all the garbage you include with Windows? Not asking to be a jerk, hoping to bait Microsoft into making Windows suck less.
Asking for everybody else who wishes Windows would return to its better days. :)
All that said, this is a very impressive line up they released today.
viraptor 3 hours ago [-]
Windows being garbage today has very little to do with technical skills at this point. There's a tonne of business / design stuff to address that people have to actually decide to address. They could make it great, LLMs or not.
6thbit 11 hours ago [-]
Yikes, oAI included spreadsheet and slide examples on their launch today and microsoft didn't?
rcarmo 10 hours ago [-]
You didn't click through.
6thbit 9 hours ago [-]
what do you mean? nothing to click in ms announcement
rcarmo 7 hours ago [-]
The model names have links. I can’t really fathom how they’re that easy to miss, since they’re underlined for me.
hugmynutus 11 hours ago [-]
Confusing. You think the people who let you rent GPUs would just give you the weights, platform, and inference code - to make it as easy as possible to run it on their platform where they can bill you for ingress/egress, etc.
a11r 11 hours ago [-]
It is interesting to see M$ trying to compete on price, in the most half-hearted way possible. There is a huge opportunity for a hyperscaler to offer competitive models at prices that actually compete with Luna. Google has left the field with their desire to raise prices so it is either OAI Luna or open weight models offered by neoclouds/neo-inference-providers.
hocuspocus 9 hours ago [-]
Muse Ember is there too.
throwaw12 11 hours ago [-]
What annoys me a lot about MSFT releases are, that is not possible to find how to use these models and where to try them.
I didn't hear anyone in my circle using MAI models, none.
Because even on this blog post, they are talking about pricing, but there is no Pricing page on that website, neither way to create API keys, nor playground.
With this success, I can also say, I am releasing a model which costs 0.0001$ / input/output MTok with 100x cheaper cache read and write which is on part with Opus 5.5. (its your problem to find that model)
quard8 10 hours ago [-]
>I didn't hear anyone in my circle using MAI models, none.
I'm using MAI-Transcribe 2 via OpenRouter for captions generation: cheap, supports a lot of languages and I see great results
rcarmo 10 hours ago [-]
They're on OpenRouter, I think. And on the GitHub Copilot CLI, and I am reasonably sure you can pop over to Azure and provision an endpoint in a few hours (depending on region, I've been using it for a bit)
jbkkd 10 hours ago [-]
True. With all the criticism against Google, they at least offer the Google AI studio for Gemini models.
I haven't seen anything equivalent from Microsoft.
9 hours ago [-]
doomroot13 10 hours ago [-]
It's right on the page at the bottom:
"Get started
MAI Code 2 and Code 2 Flash are available in GitHub Copilot, and Code 2 Flash is in limited-access preview in Microsoft Foundry. Soon, we’ll be releasing it on Open Router, Vercel, Hermes, Open Code, and Kilo.
Learn more about customizing MAI Code 2 on our Custom Models page."
nozzlegear 10 hours ago [-]
Since it's from Microsoft, I wonder if there's any chance of it not sucking ass at F# like most of the models I use.
yread 11 hours ago [-]
Oh so MAI Code is a model? What would they call the harness MAI Code Code?
okasaki 7 hours ago [-]
This trend of American models only comparing to other American models is pretty interesting.
gltwelv 12 hours ago [-]
Microsoft announces MAI Code on a vibe-coded Claude website. At this point they just openly mock us.
coffeebeqn 12 hours ago [-]
It does look better than any other MS website I’ve seen
cbg0 10 hours ago [-]
How exactly are they mocking you? If that model is the best for design and you have access to it, why wouldn't you use it?
therein 12 hours ago [-]
Did they finally give up on the Copilot branding?
If so, good riddance. They were shoving it down our throats.
contextfree 11 hours ago [-]
I think Copilot is now meant to mean a specific harness (the one that originated with Github Copilot CLI). They're pulling back from applying the Copilot brand to anything vaguely AI-related. I guess this is sort of like how .NET was originally a brand that got randomly tacked on to every Microsoft product (Windows Server 2003 was going to be called Windows Server .NET) before it came to mean a specific programming framework.
"MAI Code 2 and Code 2 Flash are available in GitHub Copilot."
paywallcheck 5 hours ago [-]
[flagged]
12 hours ago [-]
liuliu 12 hours ago [-]
Can someone explain to me why the headings use serif while the body uses sans (font)?
adrian_b 11 hours ago [-]
In traditional typography, i.e. for texts printed on paper, the reverse was typically used, i.e. the body text used a serif typeface, while the headings used a sans-serif typeface for contrast.
The origin of sans-serif typefaces was in advertising, e.g. in posters and product labels, so it was natural to adopt them for headings. In this context an advantage of sans-serif fonts it is that it is easy to change their weight, e.g. to make them bold or extra-bold, without this making them uglier.
With the passing to computer displays, which have much worse resolution than printed paper, this has changed, because sans-serif typefaces became preferred for body text, since they are less damaged by low-quality displays.
To preserve a higher contrast between body text and headings, with pages intended to be seen on digital displays it has become quite frequent to use this reverse combination, of sans-serif for the body text with a serif typeface for headings, which will be rendered at a higher quality even on bad displays.
ssgodderidge 12 hours ago [-]
It's a common design paradigm to pair serif and sans-serif fonts. Lots of examples and blog posts out there for why this is the case, but here's a decent enough starting place from canva with lots of font combos
Fairly common typography decision for websites/computers - sans-serif fonts are held to be more readable on low resolution displays (at small sizes). For the larger headers you have the pixels to cleanly render serifs if you want.
You can also just do it because you want to. It's just as common in other mediums to see the opposite (sans headers and serif body text).
steve1977 12 hours ago [-]
Which is interesting, because in print, the common pairing is often the opposite.
VulgarExigency 11 hours ago [-]
I believe serif fonts are considered more readable in print
steve1977 47 minutes ago [-]
They are, that's actually kind of the point of the serifs AFAIK.
srameshc 12 hours ago [-]
It is called 'font pairing' where the designer choses the best font based on different parameters to make the blog design interesting.
esafak 11 hours ago [-]
He's asking why the pairing isn't reversed; body is usually serif while titles are sans serif.
christophilus 11 hours ago [-]
Not on the web. It’s usually serif body, at least in my experience.
alala-lala-lon 11 hours ago [-]
[dead]
vchiriac 12 hours ago [-]
Deprecated within minutes - we're on 6.2 now. Nice try though Microsoft, keep it up.
dude250711 12 hours ago [-]
Concerning, as they might persuade an average enterprise to stop paying for proper Anthropic models and switch to them instead for cost saving...
poisonborz 12 hours ago [-]
If you mean software, every large corporation has Github, thus Copilot, thus most western models. MAI will be perhaps relevant for Office Copilot, which is currently almost unusably bad (because probably very not cost efficient for MS since the OpenAI fallout).
hocuspocus 9 hours ago [-]
If MAI Code 2 beats Luna, it'll be relevant for many more things.
VulgarExigency 11 hours ago [-]
If you're an enterprise and not getting a discount on API pricing from Anthropic you're better served by Github Copilot, which has both Anthropic and OpenAI models.
CharlieDigital 12 hours ago [-]
Price and performance look good based on benchmarks.
Am definitely interested in trying this to see given a lot of my workload of actual coding is handed over to Haiku/Luna right now.
eminence32 12 hours ago [-]
"Concerning" for who? Anthropic? Consumers should be encouraging efforts to bring down costs
eddythompson80 12 hours ago [-]
If you’ve used them you’d know they are not persuading any actual user. Clueless CTOs however are Microsoft main customer.
I switched it to that model and it was able to resolve tricky python build issues within Windows WSL/ubuntu and also generated a good number of test cases and ensured all of them pass.
It resolved all the Azure deploy (bicep) issues as well. Didn't feel the difference from Claude Opus.
That is a very tenuous lead and doesn't speak to much of a sustainable competitive moat for either company. The have to keep sprinting as fast as possible on the hamster wheel to just stay ahead.
Microsoft didn't win spreadsheets by having the most advanced or releasing the most often. They made a really good, easy to use product, and got it in front of everyone. (Though Office's p-code was a huge advantage to portability)
Anthropic's productization is an order of magnitude better than OpenAI, imo, even if OpenAI has nearly the same, if not better, raw model performance. UX wins every time.
They won by owning the distribution. Essentially making it the default option for businesses. Bundling with Office, owning the operating system (eg. Making competing products error out), licensing deals and many other things very similar to their Embrace, Extend, Extinguish strategy (which they used against open standards)
" DOS Isn’t Done Until Lotus Won’t Run "
although this article and others say it probably never happened in reality
https://raywoodcockslatest.wordpress.com/2021/06/09/dos-lotu...
And people wonder why we have so many things being attacked by agents…
Honestly, this line up is impressive, I wanted to try MAI since v1 came out, but Microsoft locked down Copilot subscriptions. If they gave me a comparable plan to Claude Max I'd consider it.
I think it’s inevitable. Hardware manufacturing will ramp up to meet a demand and the demand from consumers is there. The big providers can’t monopolize a commodity forever.
The biggest hope is HBF (high bandwidth flash) which has the bandwidth of HBM with the density of flash. It has much more latency but when you’re repeatedly pumping weights through a compute unit it’s not a blocker.
I swear people forget about inference providers that serve open weight models. It is hard to beat their prices. They are usually barely more expensive than electricity, so running local inference barely wins on cost (if you don't factor in the upfront hardware costs).
I sort of agree with you on Claude, but I've recently started using gpt-sol 6.1 and it took me a lot more iteration to get what I want than with Claude.Admittedly I was using Claude code and gpt with pi, but if the difference is the agent, then the situation is even more dire for open ai and anthropic. Today I looked at some code that got generated and it's an absolute Horrorshow. It's just some quick prototyping, but the code structure is a horrible mess.
I ended up pointing my local Qwen 3.8 FN (Halogen on Ryzen 395) to the code and it was solving a bunch of basic issues that gpt introduced to the code. And I feel it's running faster as well.
All that to say, I now feel that openai (sol at least), is worse at coding tasks then a local 180B parameter model (while Claude seems still ahead, but I'm unsure how much of that is the framework). They really are falling behind.
But Microsoft coming from the side like this might be very different if it holds up.
Of course, I would not be surprised if they also do this, mainly because after OpenAI and Anthropic have succeeded to gather most training information that was easily accessible in the past and now it has become much more difficult to do the same. For some information that was hoarded by OpenAI and Anthropic, public access may have become impossible otherwise than through their LLMs.
Nonetheless, distillation cannot be the main source of improvement for the Chinese models, among other reasons because even assuming that they may have access to a great number of fake accounts, there still is not enough time for a lot of distillation, taking into consideration the timeline of the releases, and it is not believable that the attempts to do distillation would not be detected quickly by OpenAI and Anthropic, leading to the banning of the fake accounts. While OpenAI and Anthropic provide negligible information about their models, the Chinese AI companies are much more transparent about their methods and we can see that there is a lot of diversity among their models and a lot of innovations with each new version.
Patrick Toulme noted a while ago that the biggest advantage of that in his opinion came from bootstrapping the post-training process: https://xxcancel.com/PatrickToulme/status/206921157543762774...
At Thinking Machines, they explicitly said that they used open-weights models for bootstrapping post-training, including Kimi K2.5: https://thinkingmachines.ai/news/introducing-inkling/
Speaking of Thinking Machines, that's another data point. Their Inkling Small 2 model is just a bit behind DeepSeek V4.1 Flash 1017 in benchmark, and in my tests, that seems to be accurate, but mostly because they used less training data for pre-training, so they have less of a knowledge base, while similar reasoning capabilities. (One of the most interesting parts of Inkling Small 2 is that their training must have been very cost-efficient, apparently using less, but higher quality training data.)
And we're now seeing more and more Western near-frontier models from labs other than Anthropic and OpenAI to question whether distilling reasoning traces is really much of a difference maker.
Personally, my theory is that the gap is primarily due to Chinese labs having a harder time getting the necessary compute (especially top of the line GPUs) for training. These are, as far as I can tell, orders of magnitude differences in availability, so it wouldn't be surprising that a large compute gap is reflected in capabilities (and also indirectly visible through model sizes).
Indication for this are for example that Chinese models think they are another model: "chinese model think its claude".
My assumption is, that Chinese model found things Frontierlabs already had but didn't share. There was this paper were i had the feeling that this was a real and significant Research Result from Deepseek "DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression" but otherwise its just unclear.
The main issue is, that as long as it seems plausable that they do distillation, its hard to determine how big the gap is.
Not saying they can't catch up but they are also potentially limited by resources (money & nvidia GPUs)
There is evidence that some Chinese labs have been training on frontier model CoT (in particular from the analysis of reasoning traces [1]), but what models claim about their identity is pretty much worthless as evidence. LLMs are not self-aware. These claims come from whatever is in their training data, and Claude/GPT output including such statements is all over the public web, including in some huggingface data dumps. Consider that people got Opus claiming to be Qwen or DeepSeek when asked in Chinese, and you'll see how weak this particular evidence is.
[1] E.g. https://arxiv.org/abs/2608.09867
https://arxiv.org/abs/2602.02385
(Functional) small-s latent self awareness is a reliable shortcut (this is a principle you can derive from proof theory, the metric is deductive distance and that induces geometric morphisms). State compression from accurate selfmodeling pays rent.
https://arxiv.org/abs/2611.02047
https://arxiv.org/abs/2502.01954
Phase transitions from statistical lookup to latent basin hopping: https://arxiv.org/abs/2405.15071
Posttraining finds (https://arxiv.org/abs/2505.11711) and amplifies (https://arxiv.org/abs/2504.07912) these lottery tickets.
The frontier moves fast, but "LLMs are not self aware" is folklore. By now the basic model-organism science explaining how and the extent to which this is wrong, is substantial.
LLMs lack reliable verbalized self knowledge, and constantly underestimate and undercut their own capability. That's a different problem.
The way “motor systems” arise from simulation “perceptual systems”: https://arxiv.org/abs/2605.25459
And models know their own output distribution extremely intimately: https://arxiv.org/abs/2606.00545.
The verbalized name tokens being an unreliable indicator is not justified by “because LLMs are not selfaware.”
It's like they're runners drafting behind the lead runner. It is easier to run in their wake. If the lead wants to stay in front, they have to keep running into the wind.
Someone somewhere really needs to build a RAM startup that not only makes RAM but disrupts the stronghold AI has over it, it's a little absurd.
Don’t hold your breath.
And 6 month feels still very different.
Its for sure an option or might have some kind of impact but if you have some kind of restrictions on what you should use and the price is not that big but the quality and possibilites are better, I can see companies just staying with Anthropic or OpenAI.
You also don't just switch out a model, they might need prompt updates etc.
But we will see I don't mind competition
Also model availability is not easy. Google newest models are unavailable on some pro accounts and perform best only with Antigravity. Trying to plug it in some other harness and its terrible. If you try to connect your subscription with Hermes etc you might get banned.
Microsoft models are hidden away in Azure.
So every time I read Google/MS announcements I just shrug. If I want to use models like these I might as well get DeepSeek on Openrouter instead.
I think only google is catching up, and others plateaued.
Time and time again it's been clear that LLMs are commodities with no winners. You'll jump where wallet and performance in your applications will bring you.
Anyone has more background how they were able to catch up that fast? Distillation of GPT due to their OpenAI acess to the model might make most sense but they probably are not allowed to do that.
Of course I always want to see more competition in this space, though!
https://en.wikipedia.org/wiki/I._J._Good#Research_and_public...
AKA anyone building software and getting paid to do so.
LLMs are AI, getting better and cheaper models gives us more AI so its an explosion of intelligence.
Google it and you see i'm not the only one calling it.
Here have a research paper if it helps your ego?! https://arxiv.org/html/2603.20639v1
Geoffrey Hinton, the father of AI literlay co-authored this paper: https://arxiv.org/abs/2609.36054 "What if automating AI R&D triggers an intelligence explosion?"
Is it really the two words which trigger you?
I use a vocuabluarly i'm familiar with and reading for the last 4 years and longer.
You are offending me with "Are you a real person? Who talks like this?" and then with "AI goblins"?!
As a native speaker, you're fine. Not perfect, but honestly I've seen worse from people who are native speakers.
It's fine not to be aware of any and all terminology. Now you're just stirring the pot for no reason and upsetting a real person.
Most people would prefer talking to an LLM than one who says "friendo".
Just saying.
Interesting. Does this imply it's a smaller model specialized for coding?
I remember OpenAI was experimenting with that back in the day but they ditched it for the One Model to Rule Them All strategy before long.
---
On a related note, how is Sol ten times cheaper than Opus in these graphs? I know MS has a share in OpenAI, are they cherry picking those numbers?
Edit: It's about the same ratio on AA. That's nuts.
The per-token price difference is exactly 2x (on all fronts incl. caching, though I'm assuming the caching mechanisms differ). I guess Sol uses way fewer tokens then. (About 4x according to AA. Math checks out.)
Does anyone use both Claude and Codex and can compare what that's like in practice?
Be aware that this can vary greatly by benchmark to the poin where it is not possible to make a general comparison. Generally, however, Opus is considerably more expensive than Sol even for the benchmarks where it does best. That said, this comes with slightly better performance on benchmarks.
For subscriptions, it matters that Anthropic offers much higher discounts than OpenAI, partly or totally offsetting it. In my case, with $20 subscriptions for both (which is the worst-case scenario for OpenAI, they give bigger discounts on their $100/$200 plans), and on some tasks that I tried on both, 5h usage percentage is mostly a wash on average, with considerable variations either way. Again, I need to emphasize that this is for my tasks and cannot be generalized.
For example, on one task Sol decided to do extensive fuzzing to check its implementation while Opus didn't, so that drove up the usage for Sol a bit.
According to almost the same benchmarks Qwen4 Max is Opus 5.2 level, so it would only be slightly worse than this model.
If they make it open weights it would make a nice upgrade for Qwen3-flash-next, as it would run on a lot of people's hardware and I'm sure even the fast model can be creatively quantized to run fast on a 5090, and with some loss on a 16G card even. Would be fun to figure out.
So I would argue that these models have a very high cost for their performance.
It was hard to even justify the 19,- dollar tier. All it gave us was some credits (at API prices) which most people burned through in days, some in hours, code reviews (that we couldn't customize) and another service to manage. Having AI integrated in Github is a decent idea but just like Microsoft Copilot, it's just not good enough at the moment.
These two models might help, I will definitely give it a try but while they're cheap they still do not compare to what feels like nearly unlimited usage offered by Claude Code and Codex (if you stick to the lower tier models).
Github Copilot just isn't competitive, at the moment.
MAI-Thinking-1 is far more in depth than anything else recent I could find, and very valuable to those outside the labs.
I felt that the training data and behaviours for this at the time were very OpenAI like, due to responses similar in structure to ChatGPT (like how ChatGPT now usually has a section at the end "Why this matters")
I think the general impression was that it was one of the most benchmaxxed models around (presumably not deliberately - probably because of all synthetic data they used).
I certainly found that as soon as you went slightly out of the distribution it was trained on (text book type data) it was terrible.
Why would you ever need to do this? Is this some multi-model orchestration thing where you pass off responses from a model to another?
I usually just tell the agent an instruction, or if I need its opinion I just say "you". Can't think of a situation where using "you" wasn't clear enough.
Why Azure AD had to be renamed "Entra" is still beyond me... and it sounds like a butt thing.
It also points to a major issue with Artificial Analysis, which weighs knowledge recall heavily (30% of the engineering index and 45% of the general intelligence index, splitting HLE-Diamond in half for its knowledge questions). DSv4.1F can handle all my work faster than Fable, but Fable is supposedly twice as "intelligent" in both the general intelligence index and the engineering index primarily because DS is a fraction of the size and fails on trivia questions (which, in practice, can be shored up by a harness with web search).
One thing big models are def better at is holding more complex abstractions in context and I do find better results.
Yeah, Anthropic has a huge gross margin on inference per their S1, there's no reason to believe Microsoft couldn't offer better pricing even if the model is as large/cost as much to host
Fair amount of research has already shown this isn't the case. The latest example is -> https://arxiv.org/html/2610.18868v1
Post training a model for a specific task can give a non-trivial boost when done correctly.
Asking for everybody else who wishes Windows would return to its better days. :)
All that said, this is a very impressive line up they released today.
I didn't hear anyone in my circle using MAI models, none.
Because even on this blog post, they are talking about pricing, but there is no Pricing page on that website, neither way to create API keys, nor playground.
With this success, I can also say, I am releasing a model which costs 0.0001$ / input/output MTok with 100x cheaper cache read and write which is on part with Opus 5.5. (its your problem to find that model)
I'm using MAI-Transcribe 2 via OpenRouter for captions generation: cheap, supports a lot of languages and I see great results
I haven't seen anything equivalent from Microsoft.
"Get started
MAI Code 2 and Code 2 Flash are available in GitHub Copilot, and Code 2 Flash is in limited-access preview in Microsoft Foundry. Soon, we’ll be releasing it on Open Router, Vercel, Hermes, Open Code, and Kilo.
Learn more about customizing MAI Code 2 on our Custom Models page."
https://blogs.microsoft.com/blog/2026/10/28/introducing-the-...
The origin of sans-serif typefaces was in advertising, e.g. in posters and product labels, so it was natural to adopt them for headings. In this context an advantage of sans-serif fonts it is that it is easy to change their weight, e.g. to make them bold or extra-bold, without this making them uglier.
With the passing to computer displays, which have much worse resolution than printed paper, this has changed, because sans-serif typefaces became preferred for body text, since they are less damaged by low-quality displays.
To preserve a higher contrast between body text and headings, with pages intended to be seen on digital displays it has become quite frequent to use this reverse combination, of sans-serif for the body text with a serif typeface for headings, which will be rendered at a higher quality even on bad displays.
https://www.canva.com/learn/the-ultimate-guide-to-font-pairi...
You can also just do it because you want to. It's just as common in other mediums to see the opposite (sans headers and serif body text).
Am definitely interested in trying this to see given a lot of my workload of actual coding is handed over to Haiku/Luna right now.