I agree with parent, Meta has been at this a long time and its only because they have recently fallen off that they pushed this "oh give us credit its really a new org" thing. Basically, if you can't actually "win" then try to fake a restart and say we are the fastest.
Even given that, this is their second try (they had Spark 1.0). Spark 1.0 was uninteresting, this is potentially interesting, but we can't really try it yet it seems (at least not in Openrouter).
Ultimately, competition is now fierce in this broad level of intelligence/cost:
Spark 1.1, Grok 4.5, GPT 5.6 Luna, GLM 5.2
Sonnet not in the same ballpark of pricing (more expensive than Opus in many cases). Haiku has been basically abandoned.
This not being available on Openrouter really makes it hard to test. I was going to compare vs Grok 4.5 and GPT-5.6 Luna, but I don't want to deal with signing up for Meta for it unless it checks out. Please Meta make this available.
Looks like a great set of models, but there are about 20 different thinking/model levels here in this family and they are very complex to pick the right one for the task
E.g. for GeneBench Pro, it looks like you would always use GPT-5.6 Sol over Terra/Luna, its pareto optimal.
For Agents Last Exam, you would maybe want Luna, then Terra, then Luna, then Sol as you increasingly budget for tasks.
I feel that there may need to be a new auto mode in many of these cases. It selects the best model and thinking given a particular problem.
Feels like it's going to have to go that way eventually, because here we have about 20 different model and thinking levels you could use, and they're not obvious which ones are right for the given use case.
Yeah, this is most directly comparable to xAI Grok 4.5. In both cases, directionally "opus level intelligence for haiku prices" which is a really big deal for application developers who want to include models like this in their applications. I have been testing switching out haiku and sonnet for Grok 4.5, and may give this a try too (it is quite a bit cheaper, particularly for cached).
Great model, very nice. Opus class performance at Haiku level pricing (or cheaper with the token efficiency). This seems like a GLM-5.2 killer and this is what Sonnet 5 should have been.
This is a model I could really see used inside applications, where Opus or Sonnet or GPT-5.5 are too expensive.
I would really like to see a strong Deepseek v4-Flash competitor, which ideally is something like Sonnet 4.6 performance at <$0.30 per token. This is missing from main US labs.
I agree. Gemini actually is pretty good for isolated components too. But fable is much better at design than opus or gpt5.5. I have not seen as much difference elsewhere, but definitely design fable is great.
I think scaffolds and the app layer are really the two big things needed for the deployment of AI in most use cases. In general, my company says for a given problem, we prefer deterministic software as the solution first, followed by LLMs, followed by humans. That's how we approach pretty much every problem. Yes, there are many things that we do with LLMs that we can eventually get to be done with software, and many things that are done by humans that we can get to be done by LLMs.
Good article. I will only add that I think the problems outlined in the article are exacerbated by the regulatory apparatus. If this were a debate related to truth seeking alone, it would be good. But trials drive banning/unbanning of treatments and medicines, as well as their mandated coverage by insurance.
We could be much more flexible in our approach to things if we would un-ban things. For example after phase 1 trial or based on sufficient observational evidence, things can no longer be banned but have a higher standard for "insurance is now required to cover it".
I do think if you are claiming something is not permanent you should basically have to make some specific claims about the timeliness of it (eg you can use it for at least 10 years or your money back).
I really feel that many of the issues with mRNA vaccines and health studies in general are generalizations like “safe and effective”. Everything has statistical risks and benefits, and we should just share those front and center with people. Eg test results for X mean you have a Y% chance of having X, given your history and symptoms and other results. Here are low cost low risk marginal things you can do to improve statistical significance.
Similar for vaccines, just give us the numbers clearly and upfront.
This bypasses regulators from having to make claims beyond “we reviewed the data and agree with these numbers and feel that this should not be banned.” I do think it would also help to separate something “not banned” and being “required to be covered by insurance” or “required for professions like the military”. I think trying to simplify things makes things worse, because this abstraction is not real.
Really if they wanted a standout model that would really take the wind out of GLM's sails, they should have made this the new Haiku, priced at Haiku levels with this performance.
I agree with this assessment, IMO my takeaway from this is "Generally run Sonnet on low, otherwise use Opus". It's kind of like an "extra low" setting of Opus. (depends on the application for sure).
This is much more interesting of a model at $2/$10 (their launch pricing) than at full price. There are many competing models at around this level of performance.
I also like that the difference between low, medium, high, xhigh seems more spread, which is actually a good thing for people trying to tune applications. Running Sonnet 5 on low with the launch pricing makes this potentially a better fit than Haiku or open source models for some tasks. I don't think it will make sense at full price.
I don't see this as that different. Anthropic was the first one to get involved in the "AI models must be approved" regime. OpenAI just has the advantage of being second.
I've struggled with this. You definitely can have great cheap models. There are many of them open source and served profitably by neo-clouds. The big labs have basically given up on cheap models, and it is frustrating. It means applications are not likely to build as much on them anymore (we are shifting workloads from Haiku/Sonnet to Deepseek v4, for example).
I suspect the problem is that they need to charge a lot to keep revenue numbers up, and they are more worried about cannibalizing themselves than others cannibalizing them.
This is also the type of thing that makes space based data centers more viable. I was previously more skeptical on the concept but have come around.
I do think ground based centers will have better economics when they can be built though, and this addresses noise and water complaints which are the big 2 regional complaints.
It seems like lots of bottlenecks are getting solved quickly, except for maybe memory.
Following up on the Snapchat Specs launch: Its almost like this is a spite product. Like the product is designed specifically to demolish Snapchat Specs.
I know its not. Obviously Meta has been involved here a long time, but this is basically what Snapchat should probably have launched.
And these do seem cool, I would consider these for $299 even as a fun purchase.
For those not following the debate of X, there have been a surge of doctors who are saying that full body scans are net negative.
The argument is that current full body scans often have false positives, and our treatment of false positives is bad including risky biopsies. Some have gone so far as to literally state early detection does not lead to better outcomes (simply not true - genuine early detection is very helpful but what they mean is it is outweighed at the population level by false positives).
The counter argument is that more information is always good but that you must learn to handle the noise, and we should focus on improving how we handle false positives. Including, but not limited to more frequent and abundant scans of various types.
The reason this is trending is because it both includes the argument and feels nice that someone changes their mind.
Personally, I think a lot of the issues here come down to the fact that we lie about the statistics instead of just show them. A test is neither positive or negative. It’s x% updated probability that you have something.