Hacker Newsnew | past | comments | ask | show | jobs | submit | GodelNumbering's commentslogin

Gpt 6 Luna is cheaper than Deepseek 4.1 flash! Today is wild in terms of intelligence/price across the board!

it was expected no? if cost is the only reason to use oss models, they can do much better than small providers who don't have much compute.

Finally that price drop

   Prices per 1M tokens     Claude Opus 5.5    Claude Opus 5
   Cache reads              $0.20              $0.50
   Input tokens             $4                 $5
   Output tokens            $20                $25
   Cache writes             $5                 $6.25

Opus 5 is the model with highest spend on openrouter (https://openrouter.ai/rankings#task-spend) and it seems plausible that Opus 5 is/was the highest spend model in the world, and certainly Anthropic's biggest moneymaker.

If you are forced to reduce price despite raising capabilities, that certainly tells something about the market, and potentially about Anthropic future profitability too, since this model is their biggest topline contributor


I disagree - Fable melts the GPUs and they have a high incentive to move people off of that. If they have meaningfully decreased cost to serve on Opus 5.5, they can reduce prices and increase margin or at least turn off the most expensive compute.

It is only a price drop if price * tokens used is less

Happened to be testing a "review patches on a mailing list" harness I was developing; here are a sample of the latest results, testing 12 patches containing a total of 14 issues:

Opus 5.5: Found 8/14 issues. Total cost: $15.40

Fable 5.1: Found 7/14 issues. Total cost: $66.34

Opus 5: Found 6/14 issues. Total cost: $15.19

Sonnet 5: Found 2/14 issues. Total cost: $19.15

This is a relatively small sample size, but it was both the best and the cheapest.

ETA: NB this is "Equivalent API" cost as reported by claude's CLI; I was using my subscription.


UPDATE: Sorry, just noticed I typed in the Opus 5 total cost wrong -- it should be $58.19. Main point "best and cheapest" was from the actual numbers, not my typo.

Heh I’ve been doing almost the same - back testing against PR comments - and opus 5.5 matched fable 5.1. About $24 for 10 PRs.

I was surprised how much worse Astra did on correctness; I stopped testing with it. Gonna try sol and Luna but low confidence


I just told Opus 5.5 "Perform a code review on the current branch" to see what it would come up with. The results were not inspiring. It told me there were five issues, one of which was a test-coverage gap on line 848 of ProjectTemplateTests.cs. But ProjectTemplateTests.cs is only 160 lines long.

I told it that it had made a mistake in the line number, and to double-check all the line numbers. It responded "You were right to push on this: four of the five line numbers were wrong, and while checking them I found two findings that were overstated."

Then I noticed in the corner of the Claude CLI UI that it was showing "Effort: medium". I'm pretty sure I had set it to high effort before; I don't know when it reverted to medium, but that's another thing that doesn't exactly fill me with confidence.

I'll try again on high effort to see if it does better, but so far I am not impressed with Opus 5.5 on my first day of using it.


My prompts are moving in the other direction as sashiko [1], a managed pipeline developed for the Linux Kernel mailing list like a year ago. But last year's models needed a lot more structure and guidance; the results I posted are from the "single prompt" version of the same thing. The README [2] describes the difference. You can browse the contents to get an idea; basically all the prompts were actually written and iterated by Fable (and now Opus 5.5), seeing how agents failed the tests and improving them.

[1] https://github.com/sashiko-dev/sashiko

[2] https://gitlab.com/xen-project/people/gdunlap/xen-review-pro...


Good data and goes to show that Fable is melting the GPUs and is priced accordingly. I'd guess that cost to serve for Opus 5.5 is meaningfully lower through architecture advances

I doubt your test if you cant even notice that it is not the cheapest with just 4 numbers to compare.

They're claiming a drop in token use too, and that it nets to 40% cheaper.

Unfortunately, they're full of it https://artificialanalysis.ai/models/claude-opus-5-5#token-u...

It does work out to be a similar cost per task though


You should probably look at the cost/score graph by effort level instead:

https://artificialanalysis.ai/models/claude-opus-5-5#intelli...

It is most of the pareto frontier.


Not disputing the increase in quality, just stating that non-cherry-picked benchmarks show it is more verbose at Max effort

so don't use it at max? The benchmarks suggest that high/xhigh are more than sufficient to be ahead and a whole magnitude below max with regards to token usage. I'd treat that as an outlier and not how verbose the model is in general (QED I know)

You’re missing my point. I’m saying anthropic are exaggerating their results.

how are they exaggerating the results? Comparing the cost from that chart for 5 and 5.5 for medium-max effort paints a pretty clear picture:

         mean  median
 model
 5      4.135   4.245
 5.5    3.150   2.640

Again seeing how max is a clear outlier, the median cost saving is ~38%, not that far off from the proclaimed 40%.

Is verboseness the only measure of token efficiency towards overall task completion?

5.5 is higher for max effort, slightly higher for xhigh and lower for high, medium and low effort.

The biggest proportional difference seems to be at max (5.5 is 38% more) and at high (5.5 is 21% less).

I think most people run at high and xhigh. At xhigh it is close enough to be task dependent and I don't think most people will notice. At high effort I think it looks like it will be an improvement for most people.

5.5 Max should probably be compared to Fable - it performs a lot better than 5 Max.

https://artificialanalysis.ai/models/claude-opus-5-5?models=...


I don't think so, I typically use Opus 5 on High, and 5.5 scores lower on token use:

https://artificialanalysis.ai/models/claude-opus-5-5?models=...


[flagged]


Very fast you were.

Even created an account to tell us just that.

I tested it with Claude Code, and I can confirm it's way cheaper, better, faster and less verbose than Opus 5.

parent means that they could get more client / a larger part of the market, which would lead to more income (more tokens) despite lower marginal prices

Claude adapts to OpenAI’s surprising move to simply deliver better performance than Fable 5.1, better tools as well as featuring very low pricing.

Fable 5.1 literally was a money grabber. While I liked the results, tokens were burned so hard it was embarrassing, while Astra seemed to not care.

Also Claude makes it very hard to pay for additional token budgets, allowing only credit cards. I don’t use mine anymore since I don’t need it in everyday life I was dumbfounded.

So Anthropic is just copying OpenAI so to say, matching them and essentially with Opus 5.5 being Fable 5.1 in disguise, all they do is reduce costs.

Competition works.


People are paying for Opus 5? Not just burning down tokens left after they enjoyed Fable on the sub? Amazing.

My workplace doesn't even offer Fable. And on the sub, I've had a hard time understanding Opus 5, but Fable can deal with it with subagents.

If 5.5 is any better, I might try to do agentic-assisted development instead of just telling fable to delegate


Telling Fable to delegate is agentic development. At least I thought so until reading your comment.

When you need to disable data retention, you cannot use subscription plans.

Fable tends not to perform better, just cost more.

> Fable tends not to perform better, just cost more.

There are old wives' tales on how the original Fable was superb and the stuff of legend,but it as it was leaps and bounds beyond what other models were being offered then Anthropic opted replace it with a neutered version under the same name.

So today everyone can pay to use Fable, but legend has it they are paying for a nerfed replacement released under the same name.


Enterprise and most Team accounts use API pricing, they don't have an included-usage quota.

We can’t use fable at work, opus and Astra are as good as it gets.

Enterprise, and APIs

All the anti-AI people constantly say that any moment now the prices will skyrocket and in the end human work will be cheaper compared to using AI.

It doesn't look like that's happening, on the contrary the prices are falling especially when taking into account capabilities.


It's the Chinese open source models. They're barely behind the frontier, making AI a commodity, forcing openAI and Anthropic's margins downward.

I'm hardly a fan of China/Xi, but I do appreciate and benefit from this.


China doesn't care about money. Imagine a world where it's globally normalized to ask a Chinese LLM who to vote for, what happened in Hongkong, about the Uigurs, or if Taiwan is a country.

They will burn as much money as necessary to make that happen. And they have a virtually infinite amount of liquidity.


FYI the United States do not recognize Taiwan as a country.

There's only 12 countries that do on the planet, the most "relevant" of them being Guatemala and Haiti.

As or the LLM topic: you can download weights of chinese models and remove any censorship and bias. Can you do so with american closed ones?


> FYI the United States do not recognize Taiwan as a country.

For political reason, yes. Taiwan is obviously a country, though.


This is one explanation. However, if Xi Jinping believes that whoever reaches superintelligence first becomes the next global hegemon, doing this (and more, cough cough Taiwan) suddenly looks very sane solely as a way to kneecap the competition.

The goodwill/propaganda are convenient, sure, but my guess is that they aren't the primary motivation. Another possibility is that if no takeoff happens, pressuring OpenAI/Anthropic on profitability would exacerbate any damage overinvestment has done to the US stock market/economy.


The service is the value, not the model unto itself. This is where nearly all of HN is somehow entirely blind.

Capturing the users is the ad network, that's Google and OpenAI. Capturing corporate trust at a reasonable API cost, that's Anthropic's direction.

China has none of that and they never will for exactly the same reason Baidu is irrelevant globally despite being a highly capable search engine. 'Search' is also a commodity, that's not the value that Google brings to the table.

It's a search engine, anybody can build a search engine = that's what you just said.


Search is different as a service provided for free. When cost isn't in the picture, trust/convenience win.

But two products that provide essentially the same benefit, and one is significantly cheaper? Corporations maximize profit, my friend. What the model has to say about Tianammmen Square doesn't matter when we're using it to write code.


These companies are posting massive losses while also lowering prices. This sounds just like the Chinese bikeshare bubble where they were all taking massive losses in hopes that their competitor would go broke first.

In the end, everyone lost and there are millions of bikes in landfills.

If you're interested in the bikeshare bubble, Asianometry did a video on it a while ago.

https://www.youtube.com/watch?v=FQrEDq8KPiU


Anthropic is heading toward $100 billion in annual sales, at the fastest pace of any company in world history. There isn't a close second.

The notion of comparing this to Chinese bikesharing is comically absurd.



> Anthropic is heading toward $100 billion in annual sales, at the fastest pace of any company in world history. There isn't a close second.

"Allegedly".

Plus with a lot of accounting tricks, and this not being recurring to make their books look nice for the eventual IPO.


Honestly, it wouldn't surprise me if this was a conspiracy to crash the "west" AI labs. Might as well pop the AI bubble and see the USA economy go down the drain. Even if not orchestrated, I am sure they see how they could benefit from that outcome.

Now, all this talk of pacing the frontier obviously means that they are afraid of the competition. It could be the open models eating their margins, but also competing frontier models forcing them to invest more and more for diminishing returns, just to keep up. They would certainly benefit from a "Moore's Law" roadmap to pace the advances, and seeing that they lobby for US laws, it would probably mean they are more worried about increasing spending. Though outlawing both open models and Chinese models would be good for their bottom line as well.


The bike landfills from the bikeshare bubble are terrible, but I am not looking forward to the pelican landfills from the AI bubble.

You have no suspicion that this most recent collusion is in part to setup the conditions to guarantee government subsidies/bailouts?

There's zero chance of that ever happening. Pure delusion.

We haven’t been able to use opus as much as we’d want because it’s been too expensive for general use, price drop is good so I can stop juggling different models and just use this daily unless it has some weird new issues

Speaking for myself, I have not been able to use Opus as much as I’d want because its verbose prose makes human reviews of its assumptions, architecture proposals etc. more painful than its predecessors. If they’ve solved that, I’ll be accelerating through my backlog that much faster, and using tokens accordingly.

Footnote on their pricing page says:

> Cache hits and refreshes on Claude Opus 5.5 are priced at 0.05x the base input price.

If they do the same for Haiku and Sonnet 5.5 then we should also see 5c/mtok and 10c/mtok cache read for those models, respectively. Still too high for Haiku IMO, Luna is 2c/mtok.


Since something like 98% of tokens are cache hits, that's a pretty substantial drop from 0.1x

Not necessarily, if Haiku performs better than Luna.

6-Luna dropped it to 1c since you wrote this! :D

Price per task (not per token) is what really matters.

Agree. But similar to how ISP use 200 mbps (bits) instead of 25 MBPS(bytes), i think this trend isn't going away.

That analogy doesn't hold; at least w bits vs bytes it's still "data over time".

In this case it's measuring something nearly meaningless. You could charge 100 times less per token, but if task completion takes 1,000 times as many tokens, it's not much of a bargain.


60% cache read is cool, but subscription only get 25% more according to Cat. I'm confused

I think gpt 5.6 family also dropped pricing but didn't give any more usage for the subscriptions. Maybe it's a way to silently lower the value given to subscriptions while keeping API pricing competitive

25% more usage sounds about right given the other token costs are down about 20%? I don't think cache read is a big portion of the overall cost.

Anything with long context quickly gets dominated by cache reads. Especially for interactive sessions I’ve got cache read % between 95% and 98%.

In my mix it's usually 98% or 99% at which point Fable 5.1 was pretty close to the same cost as Opus 5 due to the cheaper cached read. I've seen similar numbers for other people with long-running tasks running experiment loops and than sort of thing.

> I don't think cache read is a big portion of the overall cost.

For long running tasks it is. That's what made Deepseek so cheap.


Yeah, flash models, DeepSeek, MiMo, GLM, I love those things. For simple tasks like a daily routine shit, just setting up stuff and then doing the hard stuff in Claude/Codex, that's a reasonable approach for someone like me, a "gentleman code farmer", lol. And even lower tier stuff, I have the local models taking care of. Now that Jev is out I can finally have a true AI sysadmins managing my "cloud in the basement" homelab at the cost of electricity, which is not cheap btw

source link please?

>and potentially about Anthropic future profitability too

have they ever shared anything about their revenue mix between consumer plans vs per-token billing? this is a revenue cut on their API billing, but they're not saying anything about increased limits on the plans. so all the plan revenue just got more profitable.


This is good. Probably to match GPT pricing, though frontier Claude models are still not as token efficient.

Opus 5.5 seems to consume quota a lot slower also.

"Opus 5.5 requires less compute to serve than Opus 5, and its pricing reflects that."

At this point, I am finding it extremely hard to believe that Jev team is not on a massive astroturfing campaign. This is happening all over reddit too. All LLM subreddits are getting flooded by Jev posts, many of which are made by new accounts that only talk about Jev, many obviously advertising in guise of sharing knowledge (e.g. https://www.reddit.com/r/LocalLLaMA/comments/1wn4cni/removed...)

Multiple posts on HN, including this one, are from accounts that only ever talked about Jev. Each get unusually high number of upvotes early on, enough to put them on frontpage. A multitude of commenters on such posts also seem to talk about only 1 topic.

Can all this happen organically? Yes but with vanishingly low probability, from my vantage point.


Same thing on Twitter, I noticed last night. Making extraordinary, categorical claims too.

LinkedIn, too. Tech influencers (I mean "AI Professionals") with long, rambling musings about why lames just don't get Jev.

> At this point, I am finding it extremely hard to believe that Jev team is not on a massive astroturfing campaign.

If it's an astroturfing campaign it's very likely jev powered.

Which, comes to think of it, if true, is self-reinforcing once discovered.


Not a lawyer but I can almost guarantee that the answer is yes. If it was a 'no', many malicious parties would simply start using that loophole.

In a civil contractual dispute you can only recover actual damages. If the contract was sent, and the other party performed work on it that had a cost for them, then most likely, yes, they would be awarded damages if you refused to compensate them for any costs incurred prior to notification that the acceptance had been sent in error.

The other outcome would be clearly inequitable: forcing the counter party to eat the loss for your irresponsible use of an AI agent.


But if I paste an image of your signature onto a contract that says you give me a million pounds, that can't be legally binding for you, can it?

No, because you are a separate individual who does not have authority to sign a contract on my behalf.

Claude isn't. It's a tool, that isn't capable of signing a contract any more that Adobe Acrobat or Photoshop is.

OP used it as a tool to sign the contract. The question would be whether they did so knowingly and intentionally, if not then whether that invalidates the contract.


If it isn't then it opens up the wonderful loophole where anyone can do anything and say "the AI did it" if there are negative consequences, and "I did it" if positive. Whether the AI actually did it or not. Got caught putting out a hit on someone? "Oh I just complained to Claude about the person and said yes when it asked if I'd like it to 'handle the problem', nothing more". The user must be held accountable.

> Claude isn't. It's a tool, that isn't capable of signing a contract any more that Adobe Acrobat or Photoshop is.

Incorrect. Claude.is far more capable. E.g. it found and applied the sig without user knowledge.


> Claude.is far more capable.

Claude isn't a legal entity. It is software.

OP is entirely responsible for the software they choose to use. Nobody else. If they misused that software, a court will not be sympathetic!


That's like saying a weapon is not capable of injuring.

Capable is capable - regardless of responsibility.


It's more like saying that a weapon is not capable of murdering someone.

No one is arguing that Claude can't stick an image onto a PDF. But that is not the same thing as signing a contract.


And Adobe Acrobat is far more capable than notepad. But that doesn't change the fact that they are tools, not individuals, and thus do not have the legal authority or ability to sign contracts.

I could write dozen lines of bash that finds for PDFs, pastes an image into them and emails it to someone. That doesn't mean that bash can sign contracts.


The bot had the authority - as agent. It had the ability - as the report itself shows.

> I could write dozen lines of bash that finds for PDFs, pastes an image into them and emails it to someone. That doesn't mean that bash can sign contracts

It does mean exactly that - though bash would need to get lucky with selection and positioning.


I mean in this specific case, it did, at your behest even. These contracts would probably be even easier to enforce than Claude-signed ones!

I disagree, in the same way that I'd disagree if you said the rollerball pen on my desk signed a contract.

They can be used by someone to sign a contract, but they cannot sign a contract themselves because they're just tools.


Your pen cannot sign a contract can purely because it doesn't have the capability to find a contract in your email, find a signature on your HD, and place that sig in the right place on that contract. If it did, then yes it would be capable of signing a contract just like Claude did in this case.

You are a legal entity that can be sued. Agent is a tool that you run and are fully responsible for.

No. It’s called forgery.

Wrong. Have you read your agents TOS? You run the agent, you accept all responsibility for what it does. You are free to sue Anthropic to try and get your money back but you already indemnified them of liability, so good luck.

The GP asked what happens if THEY did it, and they are (I give them the benefit of doubt, though you can’t be sure these days) a natural person. A clanker obviously can’t commit forgery – or any crime, being a nonperson – and sure, you can’t sue OpenAnthro Corp. if their clanker does anything unintended (which makes it utter lunacy that companies just blindly trust these things, but I digress), but that’s not what was asked.

But also, you can’t ToS your way out of criminal responsibility. If OpenAnthro Corp. offered services of human agents (remember those?!) and one of their agents committed a crime while working on a client request, no ToS in the world would exonerate them.


True, that would be forgery. I think proving whether the agent truly went rogue would be 'load bearing'.

If you did not yourself intentionally sign something, in sound mind, it’s obviously not binding. But ultimately it’s up to a court to decide if you’re telling the truth that you didn’t do it.

No, application of the principal of respondeat superior would most likely be applied to an AI agent the same as a human employee. An employer is held responsible for the actions of an employee even if it is clearly contrary to their intentions.

Fair point, but I’m not sure that applies to an employee literally forging the employer’s personal signature. And equating a user–computer program relationship with an employer–employee relationship (where there’s an actual contract to that effect) doesn’t feel right anyway. Agents are still just programs and programs cannot enter contracts (like employment) given that they are not persons. This is a computer system malfunctioning.

Edit: after thinking about it more, the relationship here is obviously that of a user and a tool, not an employer and an employee. Talking about employees is anthropomorphization.

In general the user of a potentially dangerous tool is criminally and civilly liable if the tool malfunctions – unless it can be shown that the tool was used and maintained correctly and the malfunction couldn’t reasonably have been predicted or prevented. And I’d say it would be rather difficult to argue at this point that LLMs doing unintended things couldn’t have been foreseen.

So I think I’ve changed my mind about this case. The user of free-range agents does definitely deserve any civil or criminal consequences of their reckless usage.


> given that they are not persons

A company isn't a "person", but it can enter into a contract with one. And a company is essentially a tool.


that is a specific legal fiction that has both historical precedent, and can and frequently is disregarded by courts based on the circumstances (the piercing of the corporate veil).

This opens obvious loophooles. If you had deleted all logs and trajectories, courts can't trace it. Law wasn't written for or has caught up to non-human entities capable of autonomously acting

it doesn't need to. the agent operates on your behalf, you are ultimately responsible.

> It feels suspicious that MiMo-V2.6 Pro gets 46 in de index while DeepSeek-V4.1 gets 39.

Why?


> W

W what?


3blue1brown did a series on this topic: https://www.youtube.com/watch?v=l6DKRf-fAAM https://www.youtube.com/watch?v=GlYgs6v2YfU (i think one more is yet to release)

Wonderful link. Just commenting to add: in the first video, Grant mentioned it would be a trilogy series.

I keep thinking that surely I missed the 3rd video in the series but no, 2 months later we are still waiting for the conclusion. I'm sure it'll be worth the wait though.

He made a 'trilogy' about probability distributions that is still missing its third video five years later. Let's hope it doesn't turn out the same!

Mimo has been one of those models that I have been rooting for since the first I used it, the 2.5 pro which I have used quite a bit, was very concise, very aware of how much context needs to be read for which tasks and would always keep the context tight. Also surprisingly good at strategic thinking. I had published a comparison between it and Terra where Terra was found to be using much more avg context for similar tasks https://dirac.run/posts/gpt-5-6-vs-mimo-2-5-pro-context-bloa...

Interesting but not surprising trend across the board seems to be, the flash models seems to have caught up with the pro-sized models of H1'26. No surprise all labs are rushing to bigger models.

EDIT: Wow, took a detailed look at the benchmarks. Mimo 2.6 pro, the 1T model leads Kimi K3, a 2.8T param model in 14 out of 15 benchmarks (and the last one is near tie)!! Good to see they also kept the price the same, and landed in the greenest quardrant of the intelligence vs speed of AA.


Every Grok release obscures their cache pricing while highlighting their input/output pricing

From their headline comparison:

  Grok: $2/$6 per million
  
  Fable: $10/$50 per million


  What this doesn't say: Grok costs 0.50/M cache read, Fable $0.25/M cache read
Long running agentic workflows are dominated by cache reads.

Just makes Grok sound deceptive, and more importantly, reliant on user's lack of understanding of costs aka predatory (which in turn is more infuriating)


muse spark 1.3 contribs cache read is $0.002 btw (~220x diff).

Is anyone treating Meta's offerings as a serious contender in any real use case? Zuck and co are burning cash hard to try to get people using their models after falling off the wagon for a couple years. It would be wild if those prices aren't total loss leaders.

Yes I am aware and I agree that it is from the IP. The capability of LLMs differ from traditional cookies and location can materially impact your experience which you may not want. If you read the conversation, I had explicit instructions to not use my location under any circumstances and it ignored the instructions

A friend of mine created GoBench[1][2] that evaluates LLMs on 9×9 Go using KataGo opponents as Elo anchors, you see real capability differences there, like Astra Max substantially leading all other models. I think strategy is a generally interesting area to evaluate LLMs on

[1] https://rolandgao.com/blog/gobench/

[2] https://rolandgao.com/gobench.pdf


A better benchmark might be asking the LLMs to write GO ai and then comparing that-- the issue is that there will be a HUGE difference in performance that depends purely on this game being in the LLM's training... but training a general LLM to directly play these games would be a waste of capacity and shouldn't be encouraged for benchmaxxing sake.

Programming an engine OTOH is a skill that is more general and they should all have.

Might be useful to have the target of the engine be some specific virtual machine that gets a strict cycle budget-- e.g. execution runs so many cycles, and result is read out of a specific memory address at the end (or when it terminates early).


This is a great idea. It could learn from its mistakes, repeat things that work, abstract complex situations. I would even want humans monitoring the project.

Excuse me? Is this thing supposed to be a borderline SGI or not? We already know LLMs are good at spitting out code.

Playing these games autoregressively isn't even the right way to use the LLM for this task (unless it was trained to do so...). It's somewhat like having a creative writing bechmark but requiring that all the input/output be base64 encoded. It can do it-- but no guarantees on the results!

And it's also just bencmaxxing bait: you can get a huge improvement on the task by RLing on it, but make no improvement on anything else. Doing so would just waste model capacity.

If you could tell that every LLM was equally not being exposed to the task then you could justify it as a test of abstract reasoning, but you can't. So it ends up on how much go transcripts ended up in the training, which is ... not a very interesting metric.


imho doing it on 9x9s kind of misses the whole point, anyone who has gotten to single digit kyu vs humans will understand when I say its effectively tic tac toe with extra steps

It's a very fun tic tac toe!

(I have not reached single digit kyu)


Don't let yourself get too comfy with 9x9! I stayed on it for too long. No way to know but I feel like it added 2 years to the time it took to get there

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: