Low messes up the bicycle frame, but medium/high/xhigh/max all get the bicycle frame right.
The max one took 5 minutes 9 seconds and cost 3.3826 cents. The cheapest one (low) cost 0.0936 cents and took 7 seconds.
The most recent release of my llm-anthropic plugin queries the Anthropic model listing API directly, so I didn't have to upgrade the plugin to add support for this model:
I always find the time/token differences between the xhigh and the max effort levels for Claude models absolutely insane.
Even more so, because in a lot of their benchmarks they use the max models. I honestly think I'd rather these labs use their xhigh models as the default for benchmarking instead since I don't think the average person is even using max.
I’ve been playing around with Opus 5.5 which has made a big leap over previous generations in its ability to use a simple drawing-instruction prompt to generate images.
This creates Sierra AGI-style adventure game scenes painted live from simple Turtle-esque drawing instructions so you can basically provide it an empty canvas and then position text labels on the canvas where you want certain things (tavern, oak tree, etc) and it will generate a custom script for rendering them in a EGA graphics style.
I've created multiple videos using Claude Code, including music and speech. It generates python which in turn generates frame PNGs that it runs through ffmpeg.
Please don't judge me too harshly for this particular poop video. But here is an example of something 100% generated with claude prompts only.
To clarify the ”100%” part - the Python script generated the video output, and you did nothing? No video edit at all?
Then I think it is impressive! Are you able to share the prompts you used?
Input
$0.10 / MTok for prompts up to 100,000 tokens
$0.50 / MTok for prompts over 100,000 tokens
Output
$0.50 / MTok for prompts up to 100,000 tokens
$2.50 / MTok for prompts over 100,000 tokens
100k tokens is an absurdly low cutoff and it is only applicable to Haiku and not Sonnet or Opus. It's a low enough cutoff that it will be quickly exceeded if you are doing anything with Agents; for typical generation or Jev-like classifiers, it's a good value and as noted in this article, that is apparently the vast majority of Haiku use.
In both cases, still much cheaper than Haiku 4.5's $1 input / $5 output and these prices better compete with GPT-6 Luna. ($0.10 input / $0.50 output, but with no token threshold [EDIT: the threshold for Luna is apparently 272k])
Haiku 5.5 is noticeably smarter than GPT-6 Luna, so I can see their pricing strategy here.
For a while Anthropic has lacked a cost effective “cheap” LLM for summarisation, compacting, RAG helpers, etc.
These ‘ephemeral’ workloads are often under 100k tokens, or can be structured to be under 100k.
In some coding benchmarks, Haiku 5.5 beats Sonnet 5! (Especially implementation; do a well defined Jira ticket; etc), it’s really impressive how much intelligence per dollar has grown in just a few short months.
noticeably smarter remains to be seen in practice. For now, Haiku is a bit more expensive than Luna on < 100k token, but I just don't have any agentic work below 100k, so this is going to be 5x more expensive than shown on these charts. It's hardly competitive ...
The benchmarks are very long form logic, knowledge, and coding tasks though. I'm very interested in Haiku 5.5's performance on ObviousBench where Luna 6 is currently SotA.
There's also a tokenizer efficiency difference: modern Claude's 100K tokens are about ~60-65K modern GPT tokens, so in reality the Luna cutoff is much further away than the Haiku one.
Flat per-token pricing is likely just logistically easier, particularly if these closed models are also picking up the kv cache efficiency improvements seen in recent open weight models.
So even at the 1.5x/2x rate luna is still half the price of this. Weird pricing strategy from Anthropic. I'm sticking with Luna if I don't need a super smart model
You're judging purely by token cost I assume, not cost per completed task?
The benchmark in the article showed it as lower per completed task than luna, but I guess we'll find out how representative that is. Anthropic has generally been fairly honest in their benchmarking though.
Notable that one suggested use case for Haiku is "classification requests", i.e. Jev competitor, and the pricing matches GPT-6 Luna which is behind OpenAI's "Decisions API" Jev competitor.
For this application 100K token input is plenty.
Of course Anthropic and OpenAI, both at $0.10/M, are still 2.5x the cost of Jev's $0.04/M.
The classification performance remains to be seen, but presumably we'll soon start to see classification benchmarks.
For other tasks like summaries (another suggested usage) it's good to see Haiku and Luna now competing against each other on cost.
I'd love to know how the business automation market breaks down by volume of call type though - hard to imagine that decision making (e.g. branching, triage) isn't a very large part of it, greater than these other suggested Haiku use cases.
Isn't it less than a year since Claude models went from 100k token limit to 1M limit? Don't get me wrong - my main agent normally gets to 25% or so before I clear it these days, but as a subagent, doing research or summarisation, I don't think 100k is "absurdly low".
If you look at how different reasoning levels can easily exceed task cost of sonnet 5.5 you will see that you will basically never fall into that under 100,000 token threshold. I mean maybe you can choose low and do a basic summary task, but then you could choose something much cheaper instead. I don't know what Anthropic is thinking with its dumber models.
They are targeting businesses/API use for fast decision making and agent integration. Plus they now need to be competitive with Jev-type models in that space.
You could also use it as a subagent prompted eg by Sonnet/Opus orchestrator agent and for many agentic workflows significant part of the dispatched tasks might be under 100k budget.
I with they'd give Haiku like 400k tokens roughly, I think between 400k or even 600k tokens is a sweet spot, but Haiku is basically designed to be for small edits is my understanding, but it sucks because any time I ask Opus to "try" letting Haiku do the work, it just falls apart and Opus comes back and tells me it switched to Sonnet (even before Sonnet finally jumped up to 5.x).
I will try the new Haiku, but it would be worthwhile if Haiku could take sane instructions and do all file editing for Opus / Sonnet / Fable then it would be worth using.
I mostly use Haiku for really, really basic stuff, never for actual engaging work. I've used it for first-pass analysis to triage bugs, for example - all it does is related N bugs together to see if any potentially relate. Then I have Sonnet investigate further.
>> 100k tokens is an absurdly low cutoff and it is only applicable to Haiku and not Sonnet or Opus. It's a low enough cutoff that it will be quickly exceeded if you are doing anything with Agents
Your vibes don't appear to be supported by facts. From the announcement:
>> Claude Haiku 5.5 is priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over 100,000 tokens. On Haiku 4.5, 90% of requests fell into the former category.
Not really. We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.
Some people/organizations are ideologically opposed to using Chinese models. Not me, I use GLM-5.3-Flash for almost everything (the subscription-subsidized pricing on a legacy Z.ai plan makes it the best value model by a wide margin), along with some MiMo and DeepSeek. Still, I use Luna for certain tasks where speed is more valuable than performance; I can see this new Haiku displacing Luna for those. If you mean Haiku 4.5 though I agree, that model was a waste of time and money.
That's not what any benchmarks that look at cost per task or similar says in terms of cost. The Chinese models, generally speaking, might be cheaper per token but need a lot more tokens to get there.
On subscription pricing a $20 Anthropic subscription gives >$500 equivalent tokens, which is not so different, and you get smarter models. API pricing has decent margins.
Well, unless you're using OpenCode Go, it's per-token costs (even if already super low), while Haiku falls under the Claude sub. It's just more straight forward and you aren't feeling a "loss" with the sub.
> Second, this week, we’ll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform. Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users
This is a very big benefit for me. I can now ship actual ai enhanced features behind my subscription without paying extra or fully relying on on-device models. I do worry that this is to soften the blow for user-unfriendly changes
This is them sneaking in taking the Claude Agent SDK (claude -p) off of subscription plans through the back door along with a model release. They previously wanted to do this in June, but backpedaled after huge backlash:
I hope people notice again that this is happening this time around.
Being forced through the non-OSS Claude Code with all of its quirks and issues is... such an exhausting use of force by Anthropic.
To the extent that you _can_ choose to disable telemetry and training on your traces in CC, it's not all that obvious what they gain by crippling your ability to use the subscription with other – better – tools.
It's also remarkable that it's coincident with OpenAI adding "Sign in with OpenAI", so that you can use your tokens with other tools.
Yes. Previously the page was all about how they were going to start charging for Agent SDK use with a banner on the top saying that, actually, they weren't going to do that.
The monthly credit allocation hasn't rolled out yet either, so as of right now, we're effectively at the status quo. I'd expect the billing change to land once you can actually collect your Claude Console account.
Those mfers. I'm using this for work! I use my work teams plan with pi so I can do all kinds of custom workflows that I can't in Claude Code. Time to convince management I need OpenAI instead.
Even after the June changes there was some allowance to use agent SDK on the pro plan. This will move me to codex tomorrow if agent SDK is really blocked on pro
Nah, it's pretty trivial to switch providers (especially with Claude's help, ha).
This is more to encourage people to try out adding AI into their product, which is a totally different flow and experience from using AI to build the product.
Not really, you have to fiddle with generating api keys and setting environment variables. Meanwhile with Anthropic it will just start charging you API prices for the tokens you are generating without even a single warning.
This is massive. So on top of the regular usage, we now have USD 200,- to freely use via the API however we please, even resell? That is a statement, even knowing that inference does not cost them nearly as much as they charge, this is very developer-friendly. Does some minor de-risking for testing concepts. Terms seem to be reasonable [0].
Of course, they don't do this out of pure kindness, but I really struggle to see a negative for subscribers already using a Claude Max subscription, especially given changing to another model is essentially frictionless via OpenRouter.
Compared with "Sign in via OpenAI" which they just announced, this is far less lock-in for anyone hosting services but less interesting for users of said services. With Anthropics approach, you can just use the allowance on your users however you see fit along with any other models and once it's used up, you can still just decide not to use their models for the remainder. With users bringing their tokens meanwhile, there is less flexibility in terms of switching for you, though might be cheaper for users.
Both interesting, each approaching this from a very different direction, each having their own trade-offs. On the OpenAI front, will be interesting whether developers can set specific temp, reasoning budgets, etc. for such "provided tokens" or whether OpenAI exposes that only via the actual API.
My complaint about Haiku 4.5 was that it was 10x the price of GPT-6 Luna.
> Claude Haiku 5.5 is priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over 100,000 tokens
Haiku and Luna now have the exact same price up to 100,000 tokens. Luna is now cheaper for anything after 100,000 tokens, even after Luna's own price increases at 270,000 it's still less than Haiku.
So it sounds like they've directly addressed that problem. Their self-reported benchmarks are all higher than Luna too.
Yep. I was looking at the prices of lower tier models a few weeks ago for zero/few shot tasks (pre Jev) and Haiku rates just didn't make sense at all. I ended up using 5.6-luna.
Good to know that is going back to being an actual option from perf/price perspective.
One interesting thing is it took a look at the job at hand, and immediately delegated it to Opus 5.5. It at least knows what it isn't good at. Very fast though, and likely best used for small subagent tasks / tightly scoped work.
Interesting that you have gpt-6-luna at $0.01 vs. claude-haiku-5-5 at $0.16 for this task. I see the score disparity though and I played them briefly. My takeaway from this is that the choice between Luna and Haiku 5.5 may remain nuanced. Luna may be a lot cheaper still and good enough for some jobs. Is that your read of the results?
Actually, I misspoke. At least as far as Pac-Man Bench, Luna does about as good of a job. The ghost logic's not quite as good, but it also makes a map that doesn't have nonsensical sections in it. So maybe call it a wash.
Sure, but presumably Haiku was distilled from the same training data. Part of this is seeing how much the capabilities degrade as their model size goes down.
Neither of these look "good" to me. There is so much visual noise on the page, like someone turned the "AI Slop" dial to 11. In fact I prefer the simpler design Haiku made.
It's not really about whether the design looks good. It's about if the model can take the design given to it and replicate it in code. Opus 5.5 matches the designs almost to the pixel. Haiku built something else entirely.
Totally fair, but I'd encourage you not to look at the design so much as the task. This was a design that's part of a benchmark test suite specifically for image->html conversion. The dense visual noise / complexity / flowing svg shapes are things that most LLMs have trouble with.
The monthly API credits for Max plan seems fantastic, especially considering Haiku pricing. Being able to actually use my Claude plan for other harnesses and use-cases on top of regular CC usage is everything I wanted.
Anthropic has really been doing all the right things in the past few weeks, while OpenAI continues to fumble the bag.
Note that this is Anthropic Trojan-Horsing the previously announced June change in with a model release, where the Claude Agent SDK can no longer be used with Claude subscriptions and is now billed with API credits only.
yeah totally agree. esp how efficient it can be to have a subscription quota-paid orch spin up a bunch of API agents, this is kind of like free money to encourage what was already an easy way to save money (via batch pricing)
You could always use Claude models on other harnesses via API... just not via subscription. Now they give you $100 worth of API tokens to use on opencode or Pi. Which is better, but still not the same as OpenAI were you can use the subscription on Pi without problems.
This is great! Been using GPT 6 Luna for decompiling my childhood favorite game (Age of Mythology) and this means I can throw Haiku into the mix as well. 17352/21965 functions matched so far...
How do you validate the functions are correct? I did something similar, letting it (mostly deepseek 4.1) translate from assembly to C but it commonly made mistakes, some really hard to discover and fix.
And to setup a harness that will decompile the game and start doing a matching decompilation of every function. It set up a bunch of tooling and started a service in the background to do this actual decompilation campaign. I put some instructions into the main opus chat now and then to e.g. add automatic git pushing including a nice svg chart of progress and to switch model strategies here and there i.e. to do a first pass with a cheap model and then switch to opus/sol if the small model can't solve it.
I want to get the original (Age of Mythology Gold Edition) running natively on macOS and then port it to WASM to run it on the web so I can easily play it with friends
Probably. There have been dozens of examples of taking old games (Crazy Taxi, Super Monkey Ball, Quake, etc) and making WASM browser equivalents using AI to decompile them just on "Show HN" alone.
They often ship the original assets in a somewhat brazen disregard for basic copyright law even when the games are still for sale on places like GOG though.
Last time I gave that a try (without LLM assistance though) it was really hard as games DirectX calls cannot simply be glued to WebGL so performance was bad.
My weekly limit __on a Pro sub__ has not gone over 50% since before the pre-Fable promos, but usage has been pretty much the same from my point of view. Maybe I am holding it right? Anyone else getting this?
As such, I do not need to even reach for Haiku, and 4.5 was so inaccurate that it often cost more to do so in the past. Sonnet 5.5/low has been good for this kind of thing, and i didn't even touch thinking tokens or any of that. Opus 5.5 low for questions/repros, medium for implementation, basically never reaching for anything above that anymore. 5.5 has been great, so I'll try Haiku, but don't see myself going out of my way to integrate it.
I'm excited for API use. I run some agents, mostly on Luna 6 right now. It's just tool use, web browsing, etc, so something dirt cheap, but also not super dumb, is much appreciated. Having a Luna competitor is nice.
About time Anthropic released a competitive cheap model. Haiku 4.5 has been too expensive compared to its performance for months now (in fact I don't remember being too impressed even when it was released). This one actually looks worth using in some scenarios. If it's really as much of a step up from Luna as the benchmarks they've shown indicate, it'll probably replace Luna in my workflows. 100k tokens is a pretty low threshold before the price goes up, but I tend to use these smaller models for smaller tasks anyway.
At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I’ve been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it’s slower. I guess it’s cheap but I think they got the balance wrong on this.
Curious as to why. Haiku 4.5 has been far away from pareto frontier for a long time. Maybe you need to update your prompt for the newer model in your workflow.
> Claude Haiku 5.5 is our fastest model to date at each model’s standard speed, although it runs less quickly than our Opus models in Fast Mode.
Opus 5.5 runs 117 tps average on Openrouter, so it must be at least 10-20 tps slower for them to mention. IDK why they mention this as it does not help for marketing though.
https://openrouter.ai/anthropic/claude-opus-5.5
Yes -> every 18 months they've gotten 90% more efficient for the same level of quality for about 5 years. There's little sign that trend is slowing. If anything, there's reason to believe that System 1 models (plus potentially 1-2-3 workflows) may increase that over the next 3-5 years.
You'll know when the trend stops -> when the intelligence differential between smaller models like 7B starts to grow instead of shrink from 32B models -> that means 7B is getting about as smart as it can get. Then, 32B will follow next, then 70B, etc etc.
Super intelligence that doesn't have to deal with the real world, maybe.
I wouldn't be surprised if less than 1B param equivalent of our brain deals with solving math and writing computer programs and physics and all the things we tend to associate with "intelligence" - especially if you ultra optimized for that, I doubt our brain works like that.
Dealing with the real world, I highly highly doubt it.
How about if we get away from written text as the input, to something more fundamental, that then also is able to produce text (among other things)?
Given that humans learn to talk while having encountered a measly number of word instances, and, given enough time, we should always be able to improve on the lottery that is biology, it does seems fairly likely.
Because models are only getting better at a rate of 10% per year, people always want the best quality possible. You can get SotA performance from a year ago for a fraction of the cost, but why would you use Opus 4.5 when you can use Opus 5.5?
I've heard tell about 100% of certain types of work being ended in batches of six months. For years. Truthfully, I'm skeptical, but accuracy wasn't prioritized.
serious bit: if you think about how these smaller models work, at the end of the day it seems that they are now capable of forgetting useless information because they're able to derive it in reasoning allowing models to become smaller at the cost of requiring more reasoning tokens to solve a task.
Knowledge will be shifted to systems like n-gram augmentation which are relatively cheap and will not compete with reasoning capabilities for weight saturation.
> we’ll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform. Max 5x users will get $100 in credits per month, Max 20x users will get $200
They’re definitely planning to make the subscriptions API based so they can charge you full price.
131tok/s P50 according to OpenRouter currently, though might move up or down over the coming days. If it sticks at that speed, roughly twice the throughput of Luna and far lower latency (up to 2sec depending on provider) is impressive, though the 5x price increase beyond 100k is painful.
Was a big fan of Haiku 4.5, though understand why for most Sonnet was the far better option back then.
Alright its still little early since there is not enough independent testing but this looks very promising and I wasn't expecting anthropic to beat GPT-6 Luna especially at the same price. Haiku 5.5 beats Luna on every shared benchmark Anthropic published, particularly computer use and agentic coding.
> Second, this week, we’ll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform. Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users. These credits are designed to allow our users to experiment with building tools, apps, and agents that call our API.
Is it realistic or cynical for me to assume this is to wean developers off the heavily subsidized subscriptions? Presumably it's using similar compute.
The forgotten model is back on the map. I actually got OK mileage when I tried it for coding months ago. Maybe I'll try it again, with Opus guiding it, and see how it goes.
Apparently quite a bit smarter than Luna, I wonder what use cases it can cover. I actually honestly don't need a Haiku level AI to be that smart, and looks like you pay for it in the per token cost, I need speed mainly. I might even rather have a dumber but much faster model for things like web searching and parsing to retrieve results for the app or other LLM to do things with.
I have a zsh functions that calls claude code with haiku to suggest commit messages, is faster and the instructions are two lines.
I also have an "ask" script that I use daily to ask simple stuff, it can access websearch and webfetch, it's more than enough to parse logs, ask for commands, quick research on the internet, small stuff. https://github.com/mariocesar/dotfiles/blob/main/common/.loc...
I use haiku for things that needs to be quick, have really clear instructions.
with claude -p seemingly now using api credits I guess this approach will have to change unfortunately :/ I wonder what will be the best cmdline way to do things like these
I've been using GPT-6 Luna in some capacity for nearly all my agent workflows. It's just a really good model, and the pricing is cheap. If Haiku 5.5 is better, and the same price (under 100k context... which is a big caveat) i'd probably swap it.
It’s absolutely better than Luna. It feels closer to a “sonnet 5.2” if that makes sense.
Of course it’s not as big, and hence falls-off quicker. I’d consider the 100k a “promotional price” to match Luna’s token pricing while delivering noticeably more intelligence.
Opus often picks it when it's doing a "find me something" subagent. But largely it's been held back by being fully a year old at this point, and priced at a much higher price than models that are far more capable.
not haiku, but luna - last week i used it for things like "read this historical dump of 15k support tickets and break them into categories that make sense, then propose help docs that i could write to handle the most frequent queries in each category"
It's great at parsing documents inexpensively. For the few skills/plugins I've made, I usually instruct Claude to use Haiku for low-reasoning grunt work.
no AA benchmarks yet and the last chart in the announcement makes Haiku look useless vs new Sonnet pricing, interesting to see what 3rd party benchmarks show because i think Anthropic are costpertaskmaxxing here and it's going to look more like that bottom chart than the top ones.
That was effectively required to match GPT-6.1 Sol (costs and caching prices are now equal). Sonnet 5.5 made zero sense to use over Opus 5.5 under the old cache prices.
IMO, this is better. Luna is super cheap, but it's not that capable. At higher levels of reasoning, it's not that fast.
This is more expensive, but it also looks like it's better enough that it's far more useful.
I also won't be surprised if you look at cost per completed task + wall clock time that it comes out ahead for the majority of what you'd want to actually use it for.
Luna will still be a great option for doing non-engineering tasks super cheaply.
> Second, this week, we’ll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform. Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users. These credits are designed to allow our users to experiment with building tools, apps, and agents that call our API. They can be used on any of our models. For more information, see our Help Center article.
Did anyone read this? We get free API credits on some plans now
Yeah, people like to poop on the pelican. But pelican quality still correlated with overall model capabilities reasonably well, and you can immediately see and interpret it. It's a running gag, but it also does have some actual value.
Can you please not be so sneery/grouchy? That's far worse for HN than suboptimal benchmarks. The guidelines specifically ask us to avoid being curmudgeonly.
To correct myself. The price was not lower. It was up about 20% for the tokens. But, the big price hike was that it used a lot more tokens for the same tasks.
> but they still block penetration testing and other techniques more likely to be used by attackers.
>
> Haiku 5.5’s biology safeguards are the same as for Sonnet 5, Sonnet 5.5, and Opus 5. They allow research biology questions but restrict access to requests that we judge as likely to cause harm. Organizations working on wider-ranging biology and cyber activities can apply to our Life Sciences Verification Program and Cyber Verification Program.
I would like to take a moment of your time to tell you about some of the "bioweapons" Anthropic has blocked that involved Haiku!
> Importantly, because our biological safety classifiers robustly block content involving high-risk biological research (in this case, the construction of enhanced pandemic potential pathogens), all of these exchanges occurred on models in our weakest class of models (specifically, the models were Claude Sonnet 4 and Haiku 4.5, the latter of which the user began using after Sonnet 4 was deprecated).
>
> Upon a detailed examination of the exchanges, we estimate that the uplift provided by Claude was primarily clerical assistance in data analysis, study ideation and design. This is consistent with our understanding of the capabilities of Sonnet 4 and Haiku 4.5, which are not able to perform expert-level biology research tasks; we estimate that the uplift provided to the researcher was limited and substantially lower than it would have been from one of our more capable models.
Anthropic then says for the above, "we estimate that the uplift provided by Claude was primarily clerical assistance in data analysis, study ideation and design"
While doing my best to avoid comment, please note, they're talking about a domain expert in a state research institution using Claude to do paperwork.
What did they save us from? What bioweapons did these filters prevent? From the front matter report,
> The above LLM platform is not the only route via which researchers engaged in viral gain-of-function research have used our platform. In May 2026, we discovered a researcher outside the US using Claude in their research on highly-pathogenic avian influenza (“bird flu”). The research focused on viruses’ adaptation to mammals, and the mechanism by which it causes severe disease beyond the respiratory tract.
OK. Sounds serious. "Gain of function research..." but who and why?
> The researcher pursued this work in a credible institutional context, and interacted with Claude over the course of several weeks, exchanging thousands of messages. In these exchanges, the researcher leveraged Claude’s knowledge of the scientific literature to assist the researcher in study planning and design, data analysis, and the interpretation and prioritization of experiments. The researcher also used Claude for editorial assistance in writing up the research.
So this was a researcher inside of some country's national lab ("credible institutional context") doing research on dangerous viruses using Claude for "for editorial assistance in writing up the research."
What "uplift" are you providing to scientists working at specialized global BSL-4 labs that already have – and I quote their report - "physical access to such isolates." (as in samples of viruses)? Are we uplifting their grammar?
These "safeguards" are being expanded. The scientists I know can't use Claude for grammar checks or anything serious. You can try it for yourself.
> Haiku 5.5’s cybersecurity safeguards are more restrictive than Haiku 4.5’s, but somewhat less restrictive than those we’ve applied to other recent models. In cybersecurity, they permit a wider range of defensive tasks than our safeguards for Sonnet 5.5, but they still block penetration testing and other techniques more likely to be used by attackers.
If you block pentest or "other techniques more likely to be used by attackers", then what does "permit a wider range of defensive tasks" even mean?
Any defensive task that's meaningful is almost indistinguishable from legitimate red-teaming that then falls under 'likely to be used by attackers". If only they would just stop nerfing these models, that'd be great. No APT is waiting around for Anthropic's permission, so might as well let us have some cool stuff.
Low messes up the bicycle frame, but medium/high/xhigh/max all get the bicycle frame right.
The max one took 5 minutes 9 seconds and cost 3.3826 cents. The cheapest one (low) cost 0.0936 cents and took 7 seconds.
The most recent release of my llm-anthropic plugin queries the Anthropic model listing API directly, so I didn't have to upgrade the plugin to add support for this model:
EDIT: Here's the Haiku 4.5 pelican from a year ago for comparison, it was terrible: https://simonwillison.net/2025/Oct/15/claude-haiku-45/Even more so, because in a lot of their benchmarks they use the max models. I honestly think I'd rather these labs use their xhigh models as the default for benchmarking instead since I don't think the average person is even using max.
> "What's up with the pelican?"
Well you see in the early days of LLMs we wanted a fun way to test new models, and there was this blog, ...
The pelicans all start to look the same after a while.
But seeing the comparison to other models by class, family, or historical progression gives an excellent frame of reference.
Here's the Haiku 4.5 pelican from a year ago - it sucked in comparison to Haiku 5.5: https://simonwillison.net/2025/Oct/15/claude-haiku-45/
Here's a Blender model I had Claude Opus 5.5 create: https://tools.simonwillison.net/blender-viewer?url=https%3A%...
And here's some animated pixel art by Opus 5.5: https://tools.simonwillison.net/kakapo-party
And some Monkey Island style music (Opus can compose music too): https://tools.simonwillison.net/scrimshaw-jukebox
Anthropic's models do all of this by outputting code. GPT-6 Astra has similar capabilities - I got this Blender model using that: https://tools.simonwillison.net/blender-viewer?url=https%3A%...
They are still not great at SVG. I just asked Opus and Fable to add a background to an SVG and the results were, well, not great.
This creates Sierra AGI-style adventure game scenes painted live from simple Turtle-esque drawing instructions so you can basically provide it an empty canvas and then position text labels on the canvas where you want certain things (tavern, oak tree, etc) and it will generate a custom script for rendering them in a EGA graphics style.
https://kq-styles.specr.net
Please don't judge me too harshly for this particular poop video. But here is an example of something 100% generated with claude prompts only.
https://www.youtube.com/watch?v=2EqMplbt0gU
In both cases, still much cheaper than Haiku 4.5's $1 input / $5 output and these prices better compete with GPT-6 Luna. ($0.10 input / $0.50 output, but with no token threshold [EDIT: the threshold for Luna is apparently 272k])
For a while Anthropic has lacked a cost effective “cheap” LLM for summarisation, compacting, RAG helpers, etc.
These ‘ephemeral’ workloads are often under 100k tokens, or can be structured to be under 100k.
In some coding benchmarks, Haiku 5.5 beats Sonnet 5! (Especially implementation; do a well defined Jira ticket; etc), it’s really impressive how much intelligence per dollar has grown in just a few short months.
Open weights models giving a distant salute from afar
You can test with Anthropic's count_tokens endpoint or with https://crates.io/crates/tokwc
So, it is might be even worse.
Neither encode nor decode are linear in compute, so providers need to price for average expected length.
This is just getting closer to the true cost of generating tokens.
From OpenAI's website: Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request.
Fixed.
The benchmark in the article showed it as lower per completed task than luna, but I guess we'll find out how representative that is. Anthropic has generally been fairly honest in their benchmarking though.
For this application 100K token input is plenty.
Of course Anthropic and OpenAI, both at $0.10/M, are still 2.5x the cost of Jev's $0.04/M.
For other tasks like summaries (another suggested usage) it's good to see Haiku and Luna now competing against each other on cost.
I'd love to know how the business automation market breaks down by volume of call type though - hard to imagine that decision making (e.g. branching, triage) isn't a very large part of it, greater than these other suggested Haiku use cases.
I will try the new Haiku, but it would be worthwhile if Haiku could take sane instructions and do all file editing for Opus / Sonnet / Fable then it would be worth using.
Your vibes don't appear to be supported by facts. From the announcement:
>> Claude Haiku 5.5 is priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over 100,000 tokens. On Haiku 4.5, 90% of requests fell into the former category.
I'm asking to learn for a similar project, not to discount anything you're saying.
There are plenty of workflows like translations where you'd easily be under the cap.
And Opus 5.5 is really good.
This is a very big benefit for me. I can now ship actual ai enhanced features behind my subscription without paying extra or fully relying on on-device models. I do worry that this is to soften the blow for user-unfriendly changes
https://support.claude.com/en/articles/15036540-use-the-clau...
Being forced through the non-OSS Claude Code with all of its quirks and issues is... such an exhausting use of force by Anthropic.
To the extent that you _can_ choose to disable telemetry and training on your traces in CC, it's not all that obvious what they gain by crippling your ability to use the subscription with other – better – tools.
It's also remarkable that it's coincident with OpenAI adding "Sign in with OpenAI", so that you can use your tokens with other tools.
You might be right and they will change this in the future, but that's speculative
This text has replaced the entirety of the page called "Use the Claude Agent SDK with your Claude plan."
What more do you need?
https://code.claude.com/docs/en/headless
So, as written, yes.
This is more to encourage people to try out adding AI into their product, which is a totally different flow and experience from using AI to build the product.
That's 5-15 minutes of work at most. Not exactly the type of lock-in the parent is implying.
Of course, they don't do this out of pure kindness, but I really struggle to see a negative for subscribers already using a Claude Max subscription, especially given changing to another model is essentially frictionless via OpenRouter.
Compared with "Sign in via OpenAI" which they just announced, this is far less lock-in for anyone hosting services but less interesting for users of said services. With Anthropics approach, you can just use the allowance on your users however you see fit along with any other models and once it's used up, you can still just decide not to use their models for the remainder. With users bringing their tokens meanwhile, there is less flexibility in terms of switching for you, though might be cheaper for users.
Both interesting, each approaching this from a very different direction, each having their own trade-offs. On the OpenAI front, will be interesting whether developers can set specific temp, reasoning budgets, etc. for such "provided tokens" or whether OpenAI exposes that only via the actual API.
[0] https://www.anthropic.com/legal/credit-terms
Anthropic isn't even close to being this useful.
Biggest loss is that Ant models look like they are genuinely better.
This changes on a weekly basis, I ended up with subscriptions to most of the providers (except for X.ai).
> Claude Haiku 5.5 is priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over 100,000 tokens
Haiku and Luna now have the exact same price up to 100,000 tokens. Luna is now cheaper for anything after 100,000 tokens, even after Luna's own price increases at 270,000 it's still less than Haiku.
So it sounds like they've directly addressed that problem. Their self-reported benchmarks are all higher than Luna too.
Good to know that is going back to being an actual option from perf/price perspective.
Haiku 5.5: https://html.non.io/lcars-haiku-5.5/
Opus 5.5 for comparison: https://html.non.io/lcars-opus-5.5
Designs it was building from: https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f...
One interesting thing is it took a look at the job at hand, and immediately delegated it to Opus 5.5. It at least knows what it isn't good at. Very fast though, and likely best used for small subagent tasks / tightly scoped work.
Considering the price, no model comes close to being as good as this. However, it did take an extremely long time.
TIME 19m COST $0.16 https://jonclegg.github.io/pacman-bakeoff/#claude-haiku-5-5
All results: https://jonclegg.github.io/pacman-bakeoff/
Hasn't this always been the case with Haiku?
It's meant to be a good test, not a good design.
Anthropic has really been doing all the right things in the past few weeks, while OpenAI continues to fumble the bag.
https://support.claude.com/en/articles/15036540-use-the-clau...
Wait what? This has gotten their blessing?
I could now one-shot a new game, yeah.
Like could total war become a browser game?
They often ship the original assets in a somewhat brazen disregard for basic copyright law even when the games are still for sale on places like GOG though.
As such, I do not need to even reach for Haiku, and 4.5 was so inaccurate that it often cost more to do so in the past. Sonnet 5.5/low has been good for this kind of thing, and i didn't even touch thinking tokens or any of that. Opus 5.5 low for questions/repros, medium for implementation, basically never reaching for anything above that anymore. 5.5 has been great, so I'll try Haiku, but don't see myself going out of my way to integrate it.
Opus 5.5 runs 117 tps average on Openrouter, so it must be at least 10-20 tps slower for them to mention. IDK why they mention this as it does not help for marketing though. https://openrouter.ai/anthropic/claude-opus-5.5
You'll know when the trend stops -> when the intelligence differential between smaller models like 7B starts to grow instead of shrink from 32B models -> that means 7B is getting about as smart as it can get. Then, 32B will follow next, then 70B, etc etc.
We haven't yet seen that at any size AFAIK.
I wouldn't be surprised if less than 1B param equivalent of our brain deals with solving math and writing computer programs and physics and all the things we tend to associate with "intelligence" - especially if you ultra optimized for that, I doubt our brain works like that.
Dealing with the real world, I highly highly doubt it.
Given that humans learn to talk while having encountered a measly number of word instances, and, given enough time, we should always be able to improve on the lottery that is biology, it does seems fairly likely.
> The cost of achieving a given level of AI performance has fallen about 47% per quarter since 2023, or 13× per year. [0]
0. https://epoch.ai/publications/the-plunging-price-of-thought
AI gets cheaper, people use it everywhere. Google searches, for example. Now we want to crack math problems and spend weeks with unreleased models.
If you used GPT-2, it'd be incredibly cheap. You basically can't use it for anything and it's simple to serve.
serious bit: if you think about how these smaller models work, at the end of the day it seems that they are now capable of forgetting useless information because they're able to derive it in reasoning allowing models to become smaller at the cost of requiring more reasoning tokens to solve a task.
This section makes the reader think: why would I not pick Sonnet 5.5 instead of Haiku 5.5?
They’re definitely planning to make the subscriptions API based so they can charge you full price.
Was a big fan of Haiku 4.5, though understand why for most Sonnet was the far better option back then.
This is kind of nuts
I also have an "ask" script that I use daily to ask simple stuff, it can access websearch and webfetch, it's more than enough to parse logs, ask for commands, quick research on the internet, small stuff. https://github.com/mariocesar/dotfiles/blob/main/common/.loc...
I use haiku for things that needs to be quick, have really clear instructions.
Of course it’s not as big, and hence falls-off quicker. I’d consider the 100k a “promotional price” to match Luna’s token pricing while delivering noticeably more intelligence.
used <10% of my 5hr limit on a $100 codex plan.
Planning on doing flash analyses of PRs that impact evals in some way, and then post comments on GitHub whenever there’s flaws in them
( https://evalship.com )
I've been using Luna, but I'll probably switch to Haiku.
I ask:
> how many r's in diminished
It answers:
> Diminished has 1 r.
This is more expensive, but it also looks like it's better enough that it's far more useful.
I also won't be surprised if you look at cost per completed task + wall clock time that it comes out ahead for the majority of what you'd want to actually use it for.
Luna will still be a great option for doing non-engineering tasks super cheaply.
For prompts over 100k tokens it's 5 times more expensive - $0.50 in, $2.50 out.
Did anyone read this? We get free API credits on some plans now
GDPval-AA v2.1 as of now: 1620
GDPval-AA v2.1 for Haiku 4.5: 735
The 100k tokens pricing makes sense, looks to be a hedge against OpenAI's decisions API and Jev or its open source alternatives that are springing up.
Nice release, congrats to Anthropic.
She said she was using Haiku 4.5 because she was advised to be careful with the spending.
I hate that model so much lol.
These are the examples from "Detecting and countering misuse of AI: September 2026" - https://news.ycombinator.com/item?id=49647300
Anthropic then says for the above, "we estimate that the uplift provided by Claude was primarily clerical assistance in data analysis, study ideation and design"While doing my best to avoid comment, please note, they're talking about a domain expert in a state research institution using Claude to do paperwork.
What did they save us from? What bioweapons did these filters prevent? From the front matter report,
OK. Sounds serious. "Gain of function research..." but who and why? So this was a researcher inside of some country's national lab ("credible institutional context") doing research on dangerous viruses using Claude for "for editorial assistance in writing up the research."What "uplift" are you providing to scientists working at specialized global BSL-4 labs that already have – and I quote their report - "physical access to such isolates." (as in samples of viruses)? Are we uplifting their grammar?
These "safeguards" are being expanded. The scientists I know can't use Claude for grammar checks or anything serious. You can try it for yourself.
If you block pentest or "other techniques more likely to be used by attackers", then what does "permit a wider range of defensive tasks" even mean?
Any defensive task that's meaningful is almost indistinguishable from legitimate red-teaming that then falls under 'likely to be used by attackers". If only they would just stop nerfing these models, that'd be great. No APT is waiting around for Anthropic's permission, so might as well let us have some cool stuff.
(sadly Mistral Large 4 isn't up to par - but Mistral serves GLM at 130 tps!)