Rendered at 02:41:08 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
tristanj 6 hours ago [-]
Meta is offering a 10x discount on input ($0.10 vs. $1.25/Mtok) and 20x discount on output ($0.20 vs. $4.25/Mtok) if you opt in to let them train on your data.
I enabled all data sharing settings but still don’t have a message about free use on that screen - the help page says free tokens are available to “some” users - is that 1% of users, 40% of users, etc?
Does your screen have the message that you are receiving free tokens?
I tried following this page, and it's certainly a lot more complex than what Meta is offering. Different price tiers, opt-in configurations, usage based availability.. I'll take the 10x discount for flipping a param switch over this all day long.
stingraycharles 57 minutes ago [-]
Seems pretty clear to me: enable it for the projects you want, and there’s a 1M / 10M token limit per day, depending on the model you use. Assuming an average context size of 100k tokens, that is 10 to 100 requests, which is not a lot. Reason enough to prefer actually paying for Meta as well.
jjcm 4 hours ago [-]
I actually really like that pricing strategy. It's very transparent
bdangubic 3 hours ago [-]
I love the idea of this pricing strategy but there is no way meta is not training on your data regardless of your monthly invoice
simonw 2 hours ago [-]
So you think the only difference between the $1.25/million token plan and the $0.10/million token plan is that you pay them more to both lie to you and breach their contractual obligation to you?
bdangubic 5 minutes ago [-]
when has Meta ever not broken their contractual obligations (I am being serious here)? are we seriously discussing/expecting any sort of privacy related to Meta?
you can pay whatever they want, they will train and use your data, I figured this is not something that should be discussed but obviously I have been mistaken...
ray_kay777 6 hours ago [-]
This makes it a very interesting alternative to Deepseek for personal work where I don't care about the training - judging by the AA benchmarks it seems like overall cost per task is similar to the new Deepseek Flash but with better benchmarks (and inbuilt vision capabilities).
martinald 42 minutes ago [-]
Also looks incredibly fast. 150tps on openrouter (nearly all deepseek providers are around the 50tps mark).
embedding-shape 2 hours ago [-]
Except with one you know they'll release the weights and architecture back to the community, with the other, it leans towards they won't do that.
GodelNumbering 6 hours ago [-]
I think that's a fair offering tbh
HDBaseT 4 hours ago [-]
Meta, please offer this on OpenRouter too (ZDR + Non-ZDR, official Meta Provider).
dan15 4 hours ago [-]
Probably to compete with DeepSeek, which AFAIK also retains data (or at least OpenRouter says they do)
dghlsakjg 3 hours ago [-]
FYI: there are providers of deepseek that offer the same or lower pricing and zero retention policies.
greyb 3 hours ago [-]
Unfortunately, none with the same caching performance as DeepSeek proper.
ForHackernews 4 hours ago [-]
DeepSeek is really crazy cheap, though, and they don't have a giant pool of other invasive personal data to correlate it with.
jofzar 4 hours ago [-]
I hate to say this and this is because I fucking despise meta. But between DeepSeek and Meta, and trust they handle the training data correctly, I trust meta.
aand16 3 hours ago [-]
What do you mean by "correctly"?
handfuloflight 2 hours ago [-]
For example, OpenCode says they have a ZDR with DeepSeek. Some of us are skeptical that's going to be properly honored. There's no way to know.
gigatexal 4 hours ago [-]
Yikes that’s compelling pricing.
deno 6 hours ago [-]
I think it's limited to US or at least EU is excluded.
WhitneyLand 6 hours ago [-]
They chose to compare against Open AI’s mid tier model Terra instead of Sol and still lost some benchmark against it.
They left Opus in and got beat in all but one benchmark.
Nothing wrong with trying to improve, but why the marketing games?
Instead of trying to say in the post you’re “closer” to frontier, first set a clear goal to beat the Chinese labs on price or performance and demonstrate it convincingly.
Then when your ready, come back and talk frontier without playing hide the model.
spmurrayzzz 4 hours ago [-]
Given the current throughput figures on OpenRouter (~180 tk/s), its likely a much smaller param count on the order of something like Luna. I think the better, more timely comparison (re: your point on Chinese labs) would be to DeepSeek-V4-Flash-0731.
It's definitely confusing from a presentation perspective, but they are somewhat coherent comparisons if you account for the inference heuristics involved.
(They could in theory be gaming the decode speeds with much larger than normal batch sizes given the TTFT is pretty high at around 8s)
sheepscreek 1 hours ago [-]
It could be that they’re pitching Meta Muse 1.2 against Terra and Opus level models. They probably consider Sol to be a level above, along with Fable.
ac29 5 hours ago [-]
> They chose to compare against Open AI’s mid tier model Terra instead of Sol and still lost some benchmark against it.
If you scroll very slightly farther there is a benchmark that includes Sol, showing it outperforming Terra (as expected) and Spark 1.2
jjice 6 hours ago [-]
While I won't take their limited benchmarks with much salt, if it actually is this close to opus, but at a third the cost, that's pretty solid. Now, Terra is pretty damn affordable too and you're right that it's suspicious that they don't put Sol in there at all.
krm01 6 hours ago [-]
We can throw benchmarks in the bin by now. Each one I've seen is heavily biased and skewed. It holds very little reliable data points (unfortunately)
deepsquirrelnet 3 hours ago [-]
If you look at papers on benchmarks, they're usually created to expose gaps in how models are trained. It should be no surprise that models get better on them over time, because you can't get better at what you don't measure.
Cherry picking the benchmarks you present is where the falsehoods lie.
lacker 5 hours ago [-]
My conclusion is the opposite. If benchmarks were meaningless, surely Meta would be able to find some benchmark that shows they are better than Sol and Fable. The fact that they can't do that tells me that benchmarks still do mean something.
villish 2 hours ago [-]
Muse 1.1 performed relatively well according to benchmarks, putting it within spitting distance of the premier models. However, based on the results I got from it and the review videos I watched, it wasn’t even close.
Opus 5 is incredible at making games. Almost like a generation better than other models from my experience. You won't see that if you just look at the popular benchmarks..
You have to test each model on your actual use case to see how well it really performs.
nrub 5 hours ago [-]
Or they spent time optimizing their model to real world problems they're facing and didn't waste time trying to game a benchmark.
brokencode 5 hours ago [-]
Or they did try to game the benchmarks and just didn’t do it well enough.
Benchmarks are one data point, not the only one, but the easiest one to compare.
nrub 3 hours ago [-]
Right, but the point is that you can't conclude that a model is necessarily bad because it's not hitting the same scores on benchmarks. I just don't agree with lacker's conclusion, because their logic doesn't seem to consider that. Scoring lower on a benchmark doesn't strictly mean they have a bad model, but it may be the case. Like you said it's one data point, but being the easiest, and obviously most gamed, means you should probably weigh them less heavily.
bradfa 6 hours ago [-]
If you got the $20 in free credits from Meta for signing up when muse-spark-1.1 was release, please note that there's now small print stating "While using free credits your content may be used for product improvement" which was not present at muse-spark-1.1 launch when the credits were given out.
If you don't mind Meta retaining your data, the "Contributor" pricing is deepseek-v4-flash-level of low, roughly 1/10th normal muse-spark API pricing currently. Attractive if you're OK with them retaining and using your data.
giancarlostoro 6 hours ago [-]
The API costs for the version of their model that feeds things back to meta is also drastically lower.
Very interesting they have a way cheaper "contributor" version "used to improve our products", how much of that is price discrimination vs the data being that valuable?
Roughly DeepSeek V4 Flash pricing, though you can get V4 from providers that don't train on your data
mchusma 7 hours ago [-]
This is a nice release and a solid improvement over Spark 1.1. It compares favorably with Grok 4.5. Not SOTA, but solid releases. I think they need to really get this more competitive with Deepseek V4 Flash / Luna pricing to move the needle.
handzhiev 6 hours ago [-]
If you are happy to share data for training, the contributor mode offers amazing price $0.10 / $0.20
mchusma 5 hours ago [-]
Yes, that is the really compelling thing here IMO. Its a viable deepseek competitor for many people, and I missed that on the first pass.
wiradikusuma 3 hours ago [-]
Hey guys I'm just wondering. Usually when someone announces a new model, they'll show you some fancy viz/video/images: "These are what my model can produce." I'm wondering if anyone is keeping track of these? Like in a gallery form, "Use this prompt to produce this output".
By itself is useful ("I want something like this, I'll just reuse the prompt and tweak"), but it can also be used as a "draw me a pelican on a bicyle" alternative. Basically feeding those prompts over model releases.
andai 2 hours ago [-]
The most interesting thing here is the kernel optimization graph.
It look like all models were still improving, when they cut off the experiment.
It reminds me of a genetic algorithm. The graph is the same: long plateaus and then massive leaps.
The only difference between the models seems to be how quickly they arrive.
wxw 6 hours ago [-]
Last I heard, everyone at Meta was using Claude Code.
Any insiders know how Muse Code is doing internally?
dxxmxnd 5 hours ago [-]
Everyone is still using claude or codex if they aren’t forced off of it. Nobody is going to use a worse tool in this culture.
baby 4 hours ago [-]
it's actually interesting that they're not being forbidden to use claude/codex, is Meta paying for it or is it personal accounts?
paxys 2 hours ago [-]
Meta lets engineers use the best tools for the job. I doubt anyone internally is going to be rushing to switch from Claude Code or Codex.
GodelNumbering 6 hours ago [-]
If there were, do you believe it would be in their interest to answer this publicly?
georgemcbay 6 hours ago [-]
> > Any insiders know how Muse Code is doing internally?
> If there were, do you believe it would be in their interest to answer this publicly?
If it were being adopted like gangbusters in their organization, sure!
So... the fact that nobody is volunteering the information is probably a valid signal of how things are actually going...
Okay, this is getting ridiculous. Were they feeling left out?
simonw 2 hours ago [-]
I'm feeling bad for Gemini. Their cyberattack felony count is currently 0!
ororroro 2 hours ago [-]
It's Irregular again. The whole industry is eating it's own tail
drivebyhooting 2 hours ago [-]
There’s no way I’m giving Zuck any of my data.
andai 2 hours ago [-]
He already has it.
One of my professors told us about the time he did a request to Facebook to send him all his data. By law they had to send it on paper. They brought it in a big truck.
All the stuff he'd deleted was still there, just with "(deleted)" next to it.
They have a lot on people without Facebook accounts though, because their tracking stuff is all over the web.
I always found it weird that Instagram gives me much better ads than Google does... Google should know much better!
sroussey 47 minutes ago [-]
Instagram knows what content you pause on, which is a huge signal.
ipsum2 7 hours ago [-]
I wonder why they didn't compare with GPT-5.6-sol, only Terra?
wmf 7 hours ago [-]
Clearly they're positioning it as a mid model.
minimaxir 7 hours ago [-]
Which is in itself a bit weird as mid models nowadays are a golden mean fallacy. Terra is much less popular than both Luna (cost-sensitive) and Sol (performance-sensitive).
Claude Sonnet is a weird exception to the mid models because Anthropic doesn't do much with Haiku and Opus is too big.
ukblewis 5 hours ago [-]
I don’t know where you get your statistics, but I love Terra and use it all of the time. It is the default fastest model in ChatGPT/Codex today. I saw today a notice saying that the model had hit capacity briefly
redox99 6 hours ago [-]
But why include Opus then?
woadwarrior01 7 hours ago [-]
Haven't you seen the kernel optimization case study at the bottom of the page? They compare against GPT-5.6 Sol and their model is worse.
logicchains 7 hours ago [-]
Presumably because it's worse than Sol, same reason they compared it to Opus 5 not Fable.
Handy-Man 7 hours ago [-]
Their bigger model is not ready - watermelon code name was still being prepared for release as of a month ago
Ive been poking with the muse code binary - seems to be written in rust, looks similar to codex but either its a very hard fork (i also see dissimilar things like config format is different, no acp, etc) or is just heavily inspired by it (more likely).
Interesting that they have separate API pricing for "we can train on your data" (whereas iirc most of the big players either make that distinction only between subscriptions and API usage, or train on everything). Wonder how it compares to Deepseek V4 Flash given that they're similar on pricing and data policy.
liviux 7 hours ago [-]
Does this muse code have any muse spark 1.2 usage included? Can't understand from the docs.
Bolwin 7 hours ago [-]
> Muse Spark 1.2 is available today in Muse Code and in Meta Model API with expanded global access
Wasn't the previous one us only? This is probably the biggest part of the post
Anyone know if muse code is open source?
sarjann 6 hours ago [-]
I do think some of features in their harness seem interesting (workers in separate worktrees at once), recovery from crashes seem interesting.
Cappybara12 6 hours ago [-]
Is this becoming a race where we have a usual flow of a company ..
AI models,
Coding agents,
image generation tools,
and more AI models ?
kcb 7 hours ago [-]
Open the weights.
king_crimson 6 hours ago [-]
Why does every AI lab feel the need to build their own coding agent…? Don’t we have more than enough already?
sroussey 30 minutes ago [-]
Own the customer relationship.
HDBaseT 4 hours ago [-]
Outputs are a little bit more deterministic if you control the harness.
It is easy to benchmark across one harness, one system prompt and extract the most performance when you control the harness.
6 hours ago [-]
conception 4 hours ago [-]
Telemetry, marketing
eugene3306 1 hours ago [-]
Do they train on their own data?
I mean, when Meta's engineer is creating some new DINOv4 or Segment Anything, with all the scaffolding around it, do they train on that?
paulkrush 7 hours ago [-]
Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. In Muse Spark 1.2, we significantly scaled up training compute on coding tasks while expanding training environment diversity. The model also maintains its strength in other key areas like general agents.
fcoury 7 hours ago [-]
Interesting, it seems like their muse code is built upon Codex CLI?
dilyevsky 4 hours ago [-]
the soak tub in the kitchen was nice
batuhandumani 3 hours ago [-]
Why should I leave Claude or GPT and switch to Meta's aMUSEment model?
AtlanticThird 6 hours ago [-]
I wish they would add a ZDR endpoint on OpenRouter
giancarlostoro 7 hours ago [-]
Will someone at Meta for the love of God make it so none of this stuff goes through Facebook.com? You want customers but most corporate firewalls block social media. Also, a lot of devs do not want their work stuff tied up to their facebook account. For the love of all things show the IG / FB logins as optional and do email as primary.
I am not a fan of Meta but I do cheer for any competitors against OpenAI and Anthropic, the duopoly is getting tiresome.
greyb 6 hours ago [-]
I honestly think they're kinda banking on piggybacking off of Facebook account integrity systems to avoid the problems that other LLM providers are facing in trying to prevent mass free trial signups for token relays and so forth.
It's not a good system obviously. Google did this as well for Gemini-CLI, but forced it to be linked to personal Google accounts (which caused a great deal of onboarding friction).
sunaookami 5 hours ago [-]
I could sign in with my Meta account that is independent and not linked to IG or Facebook. Just click "Login with Email" on dev.meta.ai.
hahahaa 6 hours ago [-]
China says hi.
giancarlostoro 4 hours ago [-]
Not in my case, I don't see any of my employers (past or current) trusting a country like China with their data.
HDBaseT 4 hours ago [-]
The decades of US brainwashing children into thinking China is the big bad guy has worked unfortunately.
The only company less trustworthy than OpenAI and Anthropic is meta.
qphe95 7 hours ago [-]
Theres no actual evidence they didn't just distill Kimi K3
polski-g 21 minutes ago [-]
There's also no evidence they didn't just distill Gemma 3. And also no evidence its not a purple popsicle.
toephu2 7 hours ago [-]
At this point, it doesn't matter who is distilling from who.
Jabrov 6 hours ago [-]
Is there any actual evidence that they did?
vcryan 6 hours ago [-]
It seems like one day, Google or Meta might produce a coding model worth discussing. That day is not today.
esafak 7 hours ago [-]
If anyone from Meta is reading, please can you publish the cost and latency for each of your benchmarks, like OpenAI does? Show us how the reasoning effort level affects them in 2D charts. This needs to become standard practice.
Readerium 6 hours ago [-]
Lol worse than DeepSeek
rvz 7 hours ago [-]
First of all, you have login to use it. Why?
After everything that you have seen with Meta, would you really trust them with a coding agent? You don't even know if your prompts are being analyzed by them on the side or if your code base is being uploaded to them. This goes for the rest of them that have closed harnesses and closed models gated by a login.
Think twice before falling for this announcement and ask yourself what they are not telling you.
minimaxir 7 hours ago [-]
Muse Spark 1.1 was released July 16th, less than a month ago. A new version release this soon (particularly after Kimi K3's release drastically overshadowed it) is a bit sus and it appears that Meta is trying a first launch do-over.
ac29 5 hours ago [-]
Doesnt seem suspect to me, training runs have checkpoints and there is no reason you cant release a checkpoint even if you are still training the model
gaogao 6 hours ago [-]
Frequent minor version bumps are pretty common these days. Opus 4.7 -> 4.8 was 42 days.
minimaxir 6 hours ago [-]
Which was in itself a do-over because Opus 4.7 received a lot of bad press on suspicion of being a regression from 4.6.
arjie 6 hours ago [-]
Somewhat surprised that Meta with all their resources couldn’t make a model that matches Composer on any frontier. All the Sparks are dominated by some other model everywhere along the frontier. Nothing fancy here since Llama defined the open model.
The use traces must be crucial to functionality which is why they’re keeping prices so low.
wmf 6 hours ago [-]
They rebooted less than one year ago so this is decent progress. Obviously users don't care about progress though.
arjie 6 hours ago [-]
Yeah, progress is useful as an internal metric, but I'm going to measure against the present frontier unfortunately. Eager to see what they come up with in the future.
https://developer.meta.com/ai/models/muse-spark/
Does your screen have the message that you are receiving free tokens?
https://platform.openai.com/settings/organization/data-contr...
Even non business accounts seem to have organization settings page access: https://platform.openai.com/settings/organization/data-contr...
But even with all sharing enabled, I’m not seeing the free tokens message there.
Do others see a free tokens message at https://platform.openai.com/settings/organization/data-contr... after enabling all sharing there?
you can pay whatever they want, they will train and use your data, I figured this is not something that should be discussed but obviously I have been mistaken...
They left Opus in and got beat in all but one benchmark.
Nothing wrong with trying to improve, but why the marketing games?
Instead of trying to say in the post you’re “closer” to frontier, first set a clear goal to beat the Chinese labs on price or performance and demonstrate it convincingly.
Then when your ready, come back and talk frontier without playing hide the model.
It's definitely confusing from a presentation perspective, but they are somewhat coherent comparisons if you account for the inference heuristics involved.
(They could in theory be gaming the decode speeds with much larger than normal batch sizes given the TTFT is pretty high at around 8s)
If you scroll very slightly farther there is a benchmark that includes Sol, showing it outperforming Terra (as expected) and Spark 1.2
Cherry picking the benchmarks you present is where the falsehoods lie.
Opus 5 is incredible at making games. Almost like a generation better than other models from my experience. You won't see that if you just look at the popular benchmarks..
You have to test each model on your actual use case to see how well it really performs.
Benchmarks are one data point, not the only one, but the easiest one to compare.
If you don't mind Meta retaining your data, the "Contributor" pricing is deepseek-v4-flash-level of low, roughly 1/10th normal muse-spark API pricing currently. Attractive if you're OK with them retaining and using your data.
Very interesting they have a way cheaper "contributor" version "used to improve our products", how much of that is price discrimination vs the data being that valuable?
Roughly DeepSeek V4 Flash pricing, though you can get V4 from providers that don't train on your data
By itself is useful ("I want something like this, I'll just reuse the prompt and tweak"), but it can also be used as a "draw me a pelican on a bicyle" alternative. Basically feeding those prompts over model releases.
It look like all models were still improving, when they cut off the experiment.
It reminds me of a genetic algorithm. The graph is the same: long plateaus and then massive leaps.
The only difference between the models seems to be how quickly they arrive.
Any insiders know how Muse Code is doing internally?
> If there were, do you believe it would be in their interest to answer this publicly?
If it were being adopted like gangbusters in their organization, sure!
So... the fact that nobody is volunteering the information is probably a valid signal of how things are actually going...
One of my professors told us about the time he did a request to Facebook to send him all his data. By law they had to send it on paper. They brought it in a big truck.
All the stuff he'd deleted was still there, just with "(deleted)" next to it.
They have a lot on people without Facebook accounts though, because their tracking stuff is all over the web.
I always found it weird that Instagram gives me much better ads than Google does... Google should know much better!
Claude Sonnet is a weird exception to the mid models because Anthropic doesn't do much with Haiku and Opus is too big.
I think it's a bit of an improvement on the Spark 1.1 pelican: https://simonwillison.net/2026/Jul/9/muse-spark-1-1/
Interesting that they have separate API pricing for "we can train on your data" (whereas iirc most of the big players either make that distinction only between subscriptions and API usage, or train on everything). Wonder how it compares to Deepseek V4 Flash given that they're similar on pricing and data policy.
Wasn't the previous one us only? This is probably the biggest part of the post
Anyone know if muse code is open source?
It is easy to benchmark across one harness, one system prompt and extract the most performance when you control the harness.
I mean, when Meta's engineer is creating some new DINOv4 or Segment Anything, with all the scaffolding around it, do they train on that?
I am not a fan of Meta but I do cheer for any competitors against OpenAI and Anthropic, the duopoly is getting tiresome.
It's not a good system obviously. Google did this as well for Gemini-CLI, but forced it to be linked to personal Google accounts (which caused a great deal of onboarding friction).
After everything that you have seen with Meta, would you really trust them with a coding agent? You don't even know if your prompts are being analyzed by them on the side or if your code base is being uploaded to them. This goes for the rest of them that have closed harnesses and closed models gated by a login.
Think twice before falling for this announcement and ask yourself what they are not telling you.
The use traces must be crucial to functionality which is why they’re keeping prices so low.