Looks like they&#x27;re still getting the post out, but the model is live now, and the system card is at <a href="https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf" rel="nofollow">https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c3...</a> .

AI reviewing AI submitted bug bounties. We have reached the dead bug bounty program theory.

Can you use AI to pre-triage the reports too?

I&#x27;ve been using Opus 4.6-4.8 in both my own and others&#x27; code to look for vulnerabilities, and I&#x27;ve found a few. I am also in the Cyber Verification Program.Fable 5 gives me policy violation errors at the moment. No idea when or if it will be fixed.

Just commenting for posterity… if this is what it claims to be, I am not looking forward to how it will empower the people who submit bug bounties to us.Historically they’ve been people from certain identifiable countries (usually developing&#x2F;poorer countries) using fuzzers with low-quality results.Now, those same people use the current-day models to good effect, but they still don’t have a true security edge and oftentimes the reports are minor or duplicative.I wonder if that’s about to deeply change.

Found this via Google:<a href="https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf" rel="nofollow">https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c3...</a>

the post is live now <a href="https:&#x2F;&#x2F;www.anthropic.com&#x2F;news&#x2F;claude-fable-5-mythos-5" rel="nofollow">https:&#x2F;&#x2F;www.anthropic.com&#x2F;news&#x2F;claude-fable-5-mythos-5</a>

I&#x27;m aware that it was marketing. I was trying to make the point that if it were really so dangerous, they wouldn&#x27;t have released it at all, (prompt injectable) &quot;safeguards&quot; or otherwise.

You fell for their fearmongering and marketing fundraising call which was done on purpose.Now they want to pause AI because of &quot;recursive self improvement&quot;.Fool me once shame on you fool me twice...

&quot;Releasing a model this capable comes with risks. Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage. We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8. To release the model both safely and quickly, we’ve tuned these safeguards conservatively—they’ll sometimes catch harmless requests, though they trigger, on average, in less than 5% of sessions. With more capable models arriving in the coming months, we’re working to improve our safeguards and reduce false positives as quickly as we can.For a small group of cyberdefenders and infrastructure providers, we’re also launching Claude Mythos 5. It’s the same underlying model as Fable 5, but with the safeguards lifted in some areas.2 Mythos 5 will initially be deployed through Project Glasswing, in collaboration with the US Government, as an upgrade to Claude Mythos Preview. It has the strongest cybersecurity capabilities of any model in the world. Soon, we intend to expand access to Mythos 5 through a broader trusted access program.&quot;

&quot;Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage. We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8.&quot;

I thought they said mythos was too dangerous to make generally available?

Maybe at this point, Fable the game will be played generated by AI as we go.

Announcement: <a href="https:&#x2F;&#x2F;www.anthropic.com&#x2F;news&#x2F;claude-fable-5-mythos-5" rel="nofollow">https:&#x2F;&#x2F;www.anthropic.com&#x2F;news&#x2F;claude-fable-5-mythos-5</a>

On face value, this feels borderline malicious.But at the same time, it&#x27;s quite funny because they seem high on their own supply. The recent communiques from claude do not pass objectivity check.And if Opus 4.6 -&gt; Opus 4.7 -&gt; Opus 4.8 is anything to go by, not sure if there are any value to their &quot;acceleration&quot;

&gt; model except to limit its effectiveness in developing frontier LLMsDoes this imply that they&#x27;re actively using it for their frontier development and that it&#x27;s very effective?

Note also that Anthropic&#x27;s definition of &quot;unsafe&quot; encompasses &quot;competing with Anthropic.&quot;In light of the ability of recent models to accelerate their own development, we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design). Using Claude to develop competing models already violates our Terms of Service, but enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms.Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT). These interventions will not affect the vast majority of coding work. We estimate they will impact ~0.03% of traffic, concentrated in fewer than 0.1% of organizations. When these interventions are active, we expect them to have minimal behavioral impact on the model except to limit its effectiveness in developing frontier LLMs. Claude will still respond helpfully to user requests. We’ll continue to improve the precision of our detection methods following the launch of this model.(From the model card document)I didn&#x27;t previously understand that they interpreted &quot;Using Claude to develop competing models&quot; so broadly. I thought that meant something like &quot;our ToS disallow distilling our models.&quot;Too bad. I&#x27;ll continue to use Claude for now, because it&#x27;s quite effective, but in the long term I don&#x27;t want powerful models like these to be controlled by any one nation or company.

I love that the conditions of getting into &quot;ultra authorized club&quot; just means that you either have deep pockets, or you&#x27;ve got the size of the audience that marketing department approves.As if being in any of these two somehow means that you won&#x27;t use the models to say, steal random people&#x27;s money.Sam Bankman-Fried or Elizabeth Holmes would have been the members of Glasswings project, if not one of the initial members. Who&#x27;s to say we don&#x27;t have similar people with access to Mythos right now?

So essentially there are 2 models, Mythos and Fable, they have the same weights but Fable is very safety-nerfed, and only ultra authorized companies have access to mythos with full capabilitiesReported benchmarks:swe-bench verified mythos 5: 95.5%; fable 5: 95.0%swe-bench pro mythos 5: 80.3%; fable 5: 80.0%terminal-bench 2.1 mythos 5: 88.0%; fable 5: 84.3%gpqa diamond mythos 5: 94.1%riemannbench mythos 5: 55.0%; mythos preview: 43.0%; opus 4.8: 34.0%arxivmath mythos 5: 78.5%critpt mythos 5: 28.6%; gpt-5.5: 27.1%; opus 4.8: 20.9%graphwalks bfs 1m mythos 5: 79.4%; mythos preview: 74.3%; opus 4.8: 68.1%humanity’s last exam mythos 5: 59.0% without tools; 64.5% with toolsbrowsecomp mythos 5: 88.0% single-agent; 93.3% multi-agentosworld-verified mythos&#x2F;fable: 85.0%gdp.pdf fable 5: 29.8% strict pass; mythos 5: 87.6% with tools on mean criteria passofficeqa pro fable 5: 57.9% on databricks’ evallegal agent benchmark mythos 5: 16.91% all-pass; 92.0% mean criterion-passhealthbench mythos 5: 62.7%healthbench professional mythos 5: 66.0%multilingual gmmlu &#x2F; milu &#x2F; include 93.2%; 92.9%; 90.5%biomysterybench 83.9% human-solvable; 46.1% human-difficultorganic chemistry mythos 5: 90.1%labbench2 patent questions mythos 5: 79.8%

&gt;Pricing for both models is $10 per million input tokens and $50 per million output tokens.

lmao, i love how the goal post is now in the &quot;multiple weeks&quot; timeline

Depends whether &quot;unable to fully automate&quot; means &quot;needs occasional human checkpoints&quot; or &quot;slowly stops caring about your actual goal.&quot; Pretty different.

It is certainly good news for those who are selling all these tokens.

Probably there will always be frontier surface which frontier model of a given generation would not be able to automate.

So in other words... the people Anthropic hired to do the R&amp;D work of training a frontier model haven&#x27;t finished training their replacement yet.

If it&#x27;s surprising to you, you haven&#x27;t used LLMs in a domain where you&#x27;re very skilled.

&gt; we believe [Mythos 5] is likely unable to fully and reliably automate R&amp;D for frontier projects spanning multiple weeksthis is good news, right? right...?

But did it mention developer in the park eating the sandwitch? That is the most important question!

The system card is 319 pages, at what point do we call it a &quot;book&quot; instead of a &quot;card&quot;?There&#x27;s a quote from a METR report on page 52:&gt;We ran [Mythos 5] on 38 of our hardest software tasks, including tasks centered around R&amp;D. [Mythos5] generally outperformed an early checkpoint of Claude Mythos Preview in these, including by succeeding on some tasks that had not been solved by any public model we have previously evaluated. However, we still observed the model occasionally failing to correctly interpret nuanced instructions in difficult tasks... Based on the available evidence, we believe [Mythos 5] is likely unable to fully and reliably automate R&amp;D for frontier projects spanning multiple weeks. We believe that a better, more confident assessment would require more time, evaluations, and information from the model developer.

Those already landed! Oh, you weren&#x27;t talking about 4.8?

It&#x27;s possible that they will transition to usage credits but why not take them at their word? To date they have continued to offer better and better models to their subscription plans.

HN needs to take a chill pill. Could it be that Mythos is expensive and they just want to give people a taste of it? I mean the alternative is not offering it at all?

I just assume Opus is constantly nerfed based on capacity. I was exclusively Claude for a long time, but the inconsistency in quality, constant outages, and slow downs were too hard to work with.I just use dumb and fast models now. I&#x27;m more engaged. I think that the higher the quality of the model, the more you tend to vibe with it, and then the more hallucinations you then miss. I&#x27;m not sure which is more productive, but I definitely burn out faster the more I vibe. At some point you&#x27;re spending your time on forums, discord, or youtube instead of engaged with what you&#x27;re building. Or you yak shave about your tooling and end up creating the 600th multi-agent gastown harness and blowing thousands of dollars on tokens to create it only to discover it&#x27;s too expense to actually use.

I agree, this looks like their plan to wane out subscriptions. This will probably come with Opus nerfs later.

More of a free trial to those authenticated and qualified with existing payment. Subscription billing is going away for sure though eventually based on the economics. Token “all you can eat” is a capital furnace otherwise.(I’m highly confident open models will eventually achieve a similar performance benchmark with distillation over time)

Get them addicted then cut them off. Oldest trick in the book.

I think they are just prioritizing enterprise customers, because this is were historically they made most money.

That&#x27;s not how it works. They don&#x27;t need revenue, they need addicts.Specifically they need businesses that fired people and adapted their business to the products, so when the unsubsidized costs hit the businesses are forced to eat the true costs.Yes they can&#x27;t afford to give the products for free, but what is essentially happening with AI services is economic dumping, keep costs artificially low to get people to fire everybody, and then Jack the rates once they have Monopoly control

That&#x27;s a big problem for all of the AI companies. Most people don&#x27;t find the technology compelling, accurate, or ethical enough to pay for a subscription.Why wouldn&#x27;t Anthropic just wait until people start subscribing, do some kind of marketing push, or obtain some kind of other sustainable revenue stream, before they go IPO? I wonder if they see the writing on the wall with all of this and want to cash out as quickly as possible?

It’s too obvious that antropic need to find way to earn enough revenue before IPO. Claude subscription isn’t earning earning much money I bet

If it&#x27;s that big of a problem to you, you&#x27;re free to just... not use the freebie?

&gt; * On June 23, we’ll remove Fable 5 from those plans. Using it after that will require usage credits. If capacity allows, we’ll extend the included window.Of course, they are a casino as well giving you free spins at the wheel with their new Fable machine, and it is done on purpose.Once there freebies have expired, many of its users will begin to gamble more on the new casino machine and will realize that it is expensive.

I don&#x27;t think they&#x27;ll phase out subscriptions ever, their whole play has been to drive demand from the bottom up. Get engineers hooked on building with claude at home, then get them to demand the ability to use it at work, and bend over their employer with no lube.They&#x27;ll probably tighten the quotas to reign in whales though.

They almost certainly already make a fuckload more money off API pricing than they do subscriptions, even if there might be more total subscription users. So offering subscriptions even at some loss is probably going to continue. Honestly, I&#x27;d be surprised if they even lost money on most subs; there are definitely Token Whales out there who mess up all the accounting up, though.Realistically I think Anthropic just has insane demand but finite capacity to run models, and Fable will just make them more money if they dedicate it to API pricing. I suspect the goal here is something like: get individual engineers&#x2F;PMs on their personal plans to taste Fable and then go to their meetings and say &quot;Yes doubling the price of every single input&#x2F;output token is a good idea, boss&quot;.

I doubt it, given the importance of those subscriptions for building and maintaining market awareness.The AI landscape is changing rapidly, and with Apple announcing the option to change the AI backend, and potential requirements enable AI choices as well, similar to EU browser choice requirements (this is more reading tea leaves than any actual requirements I am aware of). The new OS changes coming to support Googlebook, and deep Copilot&#x2F;AI integration into Windows will make maintaining user facing subscriptions essential for independent model developers like OpenAI, Anthropic, and Mistal to remain relevant longer term.If the don&#x27;t maintain that relevance there is increasing likelihood that they will get consumed by other companies whether it&#x27;s Apple, Microsoft or Google to form a foundation for their OS, or other cloud providers.

I certainly hope not. PAYG is not predictable enough for smaller companies or individuals. Where I work (non-tech company), PAYG would never fly. We aren&#x27;t big enough for that. Of course, you can set usage budgets, but there&#x27;s a pretty big difference between $200&#x2F;user&#x2F;month vs. the equivalent PAYG usage being closer to $1,000&#x2F;user&#x2F;month, if you currently use the subscription plan to its limits each week.Going PAYG only will effectively take these tools away from a huge amount of people and accelerate the push for local LLMs.OTOH, accelerating the push for local LLMs would also be fine with me.

Ooof so are we thinking that in the next 6-12 months subscriptions will be replaced with paying retail like enterprise currently?

Why would you have ethics when you could get that IPO money instead?

More ethical in some areas, actively user hostile in others

You really misunderstand what AI-doom people are worried about if you think this is anywhere near the top (or middle, or bottom) of the list of concerns.

Yup - who cares about x-risk or red lines for domestic mass surveillance anyways? I draw my red lines at prioritizing profitable customers when heavily resource constrained. That&#x27;s the true definition of evilness!

I wouldn&#x27;t call Anthropic ethical. But between Anthropic and OpenAI, Anthropic is the more ethical one

I don&#x27;t think offering a product under a certain set of terms obligates a company to maintain that offering forever. The bait and switch is certainly annoying but seeing as they&#x27;re very upfront about it you can&#x27;t say you weren&#x27;t warned. Don&#x27;t like it? Don&#x27;t use it.

It&#x27;s unethical to price it in a way not everyone can afford?

It smells like an architecture-related issue to me. They wanted to release the model asap, but they&#x27;re still implementing the fine-grained controls to constrain the model to non-subscription users.

They said they would release it back into subscriptions as capacity allows in the future. If they don&#x27;t, people are going to point back at it and rake them over the coals.

Considering their apparent nerfing of the end user plans in favor of enterprise clients, is Anthropic still the &quot;more ethical AI company&quot; like everybody loves to tell me all the time?Assuming this isn&#x27;t just a supply issue on their side, nothing says &quot;ethical AI&quot; like only allowing mega corporations to use it through cost barriers.

I don&#x27;t see how it won&#x27;t be. They lose insane amounts of money on subscription plans. I&#x27;m sure they still lose money on usage-based billing, but probably not as much.

I really don&#x27;t want this to start being the norm

Really, you have never heard of shareware or trial periods?

i have never seen this before - where you offer something and then take that away

I guess enjoy it while it lasts? OpenAI won&#x27;t be able to subsidize that forever either.

I have trouble justifying gpt after that gross stuff with the war department.Though the day is coming when there’s no distinguishing, I’m sure.

I feel like Codex made a big push to run everything on your laptop. With Claude, I get 4 cpu&#x27;s, a fair amount of ram and 30gb for every one of my dumb ideas for free in the cloud containers. Codex used to be similar, but last time I tried it just kept pushing me to run it locally on my laptop, which I really did not want to do with 20 requests going at once. That&#x27;s the main advantage for me at the moment.

I&#x27;ve only ever had the $20 month claude plan but last night took the time to setup opencode + openrouter paying for deepseek + glm. Previous experience, while extremely awkward, I&#x27;d hit my limit within one or two chat replies and it&#x27;d take me like 4 limit cycles to complete my task. Now I&#x27;m able to complete an equivalent task entire task for less than $2 in two cycles (ask -&gt; revise).I&#x27;m doing basic web development here utilizing animejs. Nothing too complicated (mostly saving time doing the scaffolding, still write the bulk of animations manually).Truly believe that American companies are going to get completely curb stomped by China due to greed, ineptitude, and violating the social contract.

Codex IME is just smarter, I think it shows given both anecdotes but also how OpenAI has always been at the front of programming competitions and math problems.But Claude models seem to be better at long term problems or more ambiguous problems.I&#x27;m curious as to what the primary benefit here. Are there secret improvements in training? There hasn&#x27;t been much in fundamental model architecture, I don&#x27;t think. What about harnesses? I wonder what&#x27;s pushing the AI. It seems like harnesses is the main thing pushing AI ever since CoT.

I&#x27;ve found Codex to be the better subscription for OpenClaw, because the limits are indeed very generous. However, I&#x27;ve found more and more that Claude Routines&#x2F;Scheduled agents can replace all the tasks I use OpenClaw for, so I&#x27;ve been slowly switching over to Claude Code. Aside from OpenClaw, I don&#x27;t find a lot of value in Codex as a harness on it&#x27;s own.

My experience is that the GPT-family of models are very smart and figure out bugs, edge cases a bit better, but it produces code that is much less mergable – if you review the code, it introduces a lot more useless&#x2F;inappropriate heavy abstractions and wrapper functions, compared to the Claude-family models which introduces the right amount of straightforward human-style code.I can recognize so much of the GPT&#x2F;Codex generated code long after it gets merged (not by me).Additionally, the time spent on every agent turn on GPT 5.5 is much longer compared to Claude Opus 4.8, which means iterating on the code takes a lot more patience, and there&#x27;s a lot more nitpicks to pick when actually using GPT 5.5 to do software engineering.Feels like GPT-style models are more geared on doing one-shot software vibing (and handling the vibe coded mixture) compared to Claude&#x27;s focus on actual software maintenance. I got a GPT Pro sub for free and wanted to cancel my Claude subscription so much, but I still keep reaching Claude models a lot more. Frustrating.

Do you use a token service like open router or just subscribe to &#x2F; unsubscribe from various models sequentially?

I have been using both codex and Claude in my day to day, trying to not get to attached to one. I want to be able to work with any provider in case one of them does something bad.

I do slightly prefer 5.5 for complex work but Claude quota usage has gotten infinitely better since the dark days a few months back - has gone from being infuriating to something I pretty much don’t have to worry about with it as a daily driver. (In fact, hitting GPT weekly quotas is more annoying now). Understand if people are still scarred by the issues + poor comms around them, though.

Wait till you kick the tires of Qwen Coder.

100% I constantly get errors and timeouts on single responses in Claude, and certainly hit limits all the time. Codex rarely. In fact, I bought a second $200 Codex plan because the quotas seemed fair and I didnt have constant issues. Claude is so great at a lot of things, but unfortunately Anthropic beats you away with a stick every chance they get.

Still satisfied with my switch to codex&#x2F;chatgpt. I couldn&#x27;t imagine switching away from claude code when it first launch but with the drastically more generous usage on codex for the same subscription tier I just can&#x27;t justify it.

It&#x27;s very disappointing but I&#x27;m assuming it&#x27;s for rational reasons on their part.

Why wouldn&#x27;t it be? How much would you pay a scientist at this point to think about a problem for you and give you a solution?

This is probably the end of &#x27;use the best model no matter the price&#x27;

The pricing can be a bit deceptive though. A good model can deliver the same results in fewer tokens.Kind of like billing a programmer by the hour.

Models are getting better, but there&#x27;s a negative change in terms of &quot;productivity&quot; per dollar. Yeah, I can throw 5 sub-agents at the problem, but the cost is getting significantly higher. And yes, I can crank out the solution much faster, but again, at some point that cost will be hard to justify. And it doesn&#x27;t matter if the cost is subsidized by a provider, if it&#x27;s paid by your company, or from your pocket. We are slowly reaching a point where the cost will be too high to justify the gains.

&gt; Pricing for both models is $10 per million input tokens and $50 per million output tokens.The step-up in intelligence looks massive (we&#x27;ll see in practice), but the price is getting to a point where it&#x27;s making me question if it&#x27;s even worth giving it a try.Good competitors will probably be out soon, which should level the playing field. I am more excited about that, just the fact that they showed that such an improvement is possible. I&#x27;m okay waiting a bit longer for this to become attainable for plebs like me.

I had a similar experience. I wanted to test it by asking it to summarise a scientific OMICs-related paper. It gave a warning about me potentially developing a bio-weapon or something like that. And switched back to Opus 4.8.

I just asked Fable for a complete code review of my lone lisp project. Started out strong. Launched Fable agents, then spent like 10 minutes thinking... And then got interrupted by a switch to Opus 4.8.&gt; Fable 5&#x27;s safety measures flagged this message for cybersecurity or biology topics.&gt; They may flag safe, normal content as well.&gt; These measures let us bring you Mythos-level capability in other areas sooner, and we&#x27;re working to refine them.Here are the results of the agentic code review session:<pre><code> ┌──────────────────────────┬───────────────┬────────────────┐
 │ Agent │ Fable 5 turns │ Opus 4.8 turns │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ values │ 134 │ 0 │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ data-intrinsics │ 104 │ 0 │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ tools-tests-build │ 81 │ 0 │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ core-intrinsics (failed) │ 25 │ 0 │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ system-memory │ 44 │ 20 │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ reader-modules │ 104 │ 25 │
 ├──────────────────────────┼───────────────┼────────────────┤
 │ linux-startup │ 95 │ 15 │
 └──────────────────────────┴───────────────┴────────────────┘
</code></pre>
This 40 minute session cost me 16% of my weekly usage. A simple code review of the most critical areas of my project got flagged as a cybersecurity risk. It really made me not want to try it again.

For me it almost immediately blocked. I had it writing code related to message digests - and it seemed to think it was too gifted for that. Gave the security warning and switched back to 4.8. Whatever... it will probably soon have the API error soon. I have mostly switched to the Codex 200 a month plan. I&#x27;ve found their 5.5 xhigh to be better than Opus 4.8 &quot;ultracode.&quot; Also, i have not once seen their servers fail for compute unavailability, unlike Anthropric which happens almost ever hour.

I expect that depends on demand, feedback, and whether GPT-6.0 gets released and is competitive

Pay-as-you-go billing is a kind of drug, I use it every now and then when I&#x27;m working on a project with Opus, in a moment you spend a fortune

Just like how Elon forced FSD in Tesla to be subscription-only (he was incentivized to do so).

&gt; The &quot;offer, then remove&quot; aspect is a bit eyebrow-raising -- it feels like they are trying to get subscribers to switch to usage-based billing, which makes me wonder if we&#x27;ll ever get it after that June 22nd window.Probably all about the IPO.

We just blocked it at our org for this reason. They will &quot;retain agent request and output data associated with this model, regardless of you Cursor Privacy Mode setting.&quot;

What does &quot;zero data retention&quot; mean? What kind of data does it need to unlock?

Fwiw it&#x27;s not available on my enterprise account: &quot;Disable zero data retention to unlock Fable 5 access&quot;

The AI circular infinite money glitch won&#x27;t last forever. I hope.If you have good expertise in a domain and access to cheaper models, you may still be more skilled than someone without expertise but a lot of money to bruteforce the problems using SOTA LLMs.

I&#x27;m about to be priced out of SOTA llms and it&#x27;s an awful feeling

Thankfully this, like most other tech, will get cheaper through the years.

This is really sad... I really didn&#x27;t want to be priced out of these models but it looks like that&#x27;s going to happen sooner rather than later.

&gt; &quot;offer, then remove&quot;Sounds like &quot;bait and wait&quot;.If you think about it, the more people pay for these new and more resource hungry models, the longer it takes for them to become no extra cost and the longer it takes the more people are tempted to pay extra.

I’m just about ready to cancel my small business 5 user plan with max licenses, because although cowork is really great.  I just find OpenAI&#x2F;Codex to be a lot better most of the time.

The point of SaaS was that the marginal cost (of supporting another user) was low. That does not apply to LLMs.

It&#x27;s interesting that we are seeing a time when subscriptions are not preferred and usage-based billing is.Pay-as-you go isn&#x27;t a common thing in SaaS. For example, except for AWS SES, all email providers are bulk-subscription based.

That’s odd, I used it on a pretty complex refactoring task and it worked for 22 mins and used only 15% of my 5-hour limit. I’m on the $200 Max plan though.

They didn&#x27;t even reset credits for this lol

The CLI when you select it says it has 2x the usage as opus. Not sure if that matches what you are seeing.I do wonder if you switched models mid-session, you would have lost all your cache. Reloading the context into cache can really eat through your usage.

they have a graph with cost comparison between the models. This model is just a little over the other models as cost. The graph is logarithmic :)

I too am on the $100 plan and I second this.I had it analyze a project I was working on with Opus 4.8, and it blew through 23% of my session limit in one go. Does not portend well for my budget.

Yes, and this is also why I haven’t yet tried the new “dynamic workflows” which spawn hundreds of agents that happily eat through your token limits.

I&#x27;m also on the $100 max plan. I let Fable rip on a complicated issue involving hot-reloading modules in a GUI app built with Racket, it&#x27;s fixed a couple issues over the last hour, and I&#x27;ve used about 17% of my session (not weekly) limit.

I would not use this if you are on a subscription. In &lt;8min it burned my entire 5hr window (which has just reset it appears, I have over 4 hours till it resets) I hadn&#x27;t used CC at all today aside from this) and then it used up ~$15 more in usage before I could stop it.I am on the $100 Max plan.

My guess is that it is a massive model similar to GPT 4.5 and $10&#x2F;$50 pricing is for its output will discourage people from using it. I also read safety = nerfed.

This is just the sales team doing their thing, applying the Law of Scarcity to drive demand.It&#x27;s the same exact speed as opus &gt;=4.5, sonnet 4.5, and twice the speed of opus &lt;=4.1It must have about the same active parameters, or else its a larger model running in turbo mode (smaller batches) and being heavily subsidized for some reason. But given most of the benchmarks are within 5% I doubt it is a much larger model. Most perplexing.

the claimed inference cost is 2x. if that is true, it is massive and remarkable that they&#x27;re able to do anything like this at all.

This serves as a good reminder that relying on AI models is borrowing your tech from someone else. They can take it away or raise the prices arbitrarily.If you rely on this as a core part of your business&#x2F;profession, you will be at their mercy and subject to whatever whims or challenges they have.

But it&#x27;s not and it&#x27;s highly disingenuous to frame it like this. Quote directly from Claude code, moments ago:&gt; Fable 5 · Most capable for your hardest and longest-running tasks · Uses your limits ~2× faster than Opus

I think the goal is &quot;private citizens: subscriptions; corporations: per-token billing.&quot; It&#x27;s getting people addicted to LLMs on cheap subscriptions so that they can then force companies to pay for expensive inference.

i doubt that&#x27;s the goal for them.  i bet they just really don&#x27;t have capacity for people using it a ton, yet they wanted people to be able to try it out while it&#x27;s new. so they compromised and made it temporarily available. and then hope they can get costs down or capacity up so they can make it more available again

Where is this text coming from?[edit] -- I see that this comes from the system card -- dang merged the comments from the other discussion so that explains the confusion.

&quot;we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design)....Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user.&quot;

Fable seems very good at finding bugs (unsurprising given Mythos lineage), so this seems a pretty smart strategy. Once you see the bugs it finds in your existing Opus code, it&#x27;s going to be hard to go back, psychologically speaking.

I switched back to Sonnet. It replies faster so I work faster. Also cheaper. But I really like the speed. I have to be more specific with what I want. Also I stop it more often than Opus. These new models will be awesome, but they need to increase the speed.

It could be my use cases, which have always seemed to be outside the wheelhouse of these models, but I find it very hard to downgrade after accessing a more capable model.Opus 4.8 produces output in 15 minutes that is 3-4 hours of my work away from output that used to take me 40ish hours (a solid week of dedicated effort).Last year(-ish, maybe it was 18 months, I forget when the jump happened), the frontier models couldn&#x27;t touch this work. The output looked like a hardworking intern on their first day. Nice formatting, decent volume of words, but no understanding.So it might work if it turns out to be a substantial leap in capability.

&gt; it feels like they are trying to get subscribers to switch to usage-based billingI think they might be hitting a point where subsidizing the expensive models for subscriptions makes less and less sense.With Opus 4.X, last month I paid 100 USD for the Max subscription and got a token equivalent of 4.1k USD.I imagine that Fable is more expensive to run.

I suspect it&#x27;ll go on the subscription plan once other providers have similar benchmarks.As annoyed as I am about this move, I get it. Users flood the newest, best model whether they really need it or not, and are efficient at using their entire quota. They&#x27;ve had so much trouble reigning in subscription usage it makes sense.

K2.6 on Cerebras is basically a preview of the future. We&#x27;ll eventually get similar performance locally with Tenstorrent hardware.

Agreed, everything since 4.6 has been worse

Kimi 2.6 has been my workhorse now. It&#x27;s as good as Opus 4.6, which, to me, was the last &quot;useful&quot; Claude model.The newer models are smarter but really ficklle and hard to get meaningful work out of4.6 was a workhorse

Enterprise subs not allowed to use Fable if they have setup zero data retention :(

One can hope it helps Claude to figure out how to solve their buggy payment system - otherwise how do I pay for these credits.

How much more clearly do they need to explain the resource constraints?If they didn&#x27;t announce it, you guys would be complaining about slowed progress.If they didn&#x27;t release it, you guys would be complaining about fake promises and marketing.If they released it without limits, the complaints would be about slow responses and outages.If they didn&#x27;t add to susbcription plans, the complaints would be about phasing out subscriptions.If they added to subscriptions with cost reflecting their resource availability, the complaints would be about how quickly it eats limits.So they choose the middle ground of providing some initial access and assessing if they can satisfy demand, only to still be ignored and accused of trying to get users hooked?We&#x27;ve already seen that they don&#x27;t have enough compute, thus the deals with SpaceX for their GPUs. It&#x27;s very reasonable that they just don&#x27;t have the capacity to support the subscription userbase on this model.

For those of us on subscription plans:* From today through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost.* On June 23, we’ll remove Fable 5 from those plans. Using it after that will require usage credits. If capacity allows, we’ll extend the included window.* After this point—when sufficient capacity allows us to do so—we aim to restore Fable 5 as a standard part of subscription plans. We intend to do this as quickly as we can.The &quot;offer, then remove&quot; aspect is a bit eyebrow-raising -- it feels like they are trying to get subscribers to switch to usage-based billing, which makes me wonder if we&#x27;ll ever get it after that June 22nd window.

&gt;Opus is too small, increase the impact of the name.Okay, how about Mythos?&gt;Increase it even more.Right, then Cosmos.&gt;Even more!Even more? Let&#x27;s try Aeon.&gt;MORE, EVEN BIGGERALRIGHT, TRY OMEGAPANTHEON 7.8 THEN

&#x2F;* What will happen first?* Anthropic runs out of genre names.* Anthropic changes the model naming convention.* AGI is achieved and handles its own naming.*&#x2F;

Of course it&#x27;s all about building the hype for the IPO :)

I wonder how Claude Fable will live up to expectations and how good those Fable&#x2F;Mythos classifiers really are. It seems a bit convenient for Anthropic to release this magical insane model when they are about to IPO.

I ran an experiment to see how far it could get with a top-down 2d game, like a more challenging version of &quot;draw a pelican.&quot; I&#x27;m waiting on Fable to rewrite the whole thing now, but I was impressed by how far Opus 4.8 got with it: <a href="https:&#x2F;&#x2F;github.com&#x2F;jmtame&#x2F;scrapland" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;jmtame&#x2F;scrapland</a>Started out as a one-shot attempt, but ~200 prompts later it&#x27;s at a place where it&#x27;s at least fun to watch the AI teams destroy each other.

That pelican better be super realistic, unreal engine 6 style graphics

I&#x27;m not familiar with model pricing trends, did they clearly state how the new pricing compares? (Note that I&#x27;m actually asking a question, and am not arguing)EDIT: Oh I see, this is the best link for pricing <a href="https:&#x2F;&#x2F;platform.claude.com&#x2F;docs&#x2F;en&#x2F;about-claude&#x2F;pricing" rel="nofollow">https:&#x2F;&#x2F;platform.claude.com&#x2F;docs&#x2F;en&#x2F;about-claude&#x2F;pricing</a>So the price is double across the board...

It&#x27;s a relatively new benchmark but from what I can tell it has serious cred behind it. I assume it will be picked up as part of the standard suite of CS-related benchmarks soon enough.

Seems like it literally popped up yesterday with the express purpose of building hype for this release.

Given it was made by cognition (team behind devin flop) who now just got to wait out until claude and gpt5 basically do all of the work for them - not very.  When you read about it, the framework is highly subjective.  Which very quickly becomes a problem because its based on heuristics that probably change a bunch with a better code model.

i worked on one of the benchmarks typically found in new model releasesthis benchmark looks very good from the methodology. a cog researcher checking the data themselves is very high signal (not scaleable so don&#x27;t take the benchmark as gospel, but directionally good)

Cognition did well in documenting their approach [1].TL;DR - they worked with OSS project maintainers to build tasks. They score models based on whether a PR is mergeable. All tasks are graded by a human researcher. SoTA models have hill-climbing to do which raises the bar and inspires confidence. I&#x27;d say it&#x27;s legit.[1]: <a href="https:&#x2F;&#x2F;x.com&#x2F;cognition&#x2F;status&#x2F;2064061031912288715" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;cognition&#x2F;status&#x2F;2064061031912288715</a>

DeepSWE is the benchmark you want to actually look out for.  Only one that aligns with actual user reported results from trying the models.

It&#x27;s an unacademic benchmark by a failed VC startup clawing for relevancy.

How credible is this benchmark? does it correlated with others real world experience?

Huh? It&#x27;s a benchmark by Cognition which (1) is building their own models and (2) offers all providers and thus has an incentive to avoid hyping up any one too much.

did they not pay them enough to get good ratings on the other 3 models?whats the logic in claiming its a borked metric when everything listed is an anthropic model.

FrontierCode is likely paid for by anthropic.

There is also the possibility that an LLM judge would be happy with some code that looks like LLM generated code. But a maintainer for a specific project might not merge it for stylistic reasons

That blog post really makes it look like it&#x27;s graded from an LLM&#x27;s estimation of an OSS maintainer&#x27;s review. I see three issues:1. That estimate could easily be wrong.2. That estimate is, of course, usable in RL training. This isn&#x27;t an inherently bad thing, and this is more or less what has improved coding models so much lately. But it does mean that other companies could and surely will do this sort of training, and Anthropic probably did too.3. OSS maintainers are far from perfect, and there&#x27;s an unfortunate uncanny valley-like effect in which a coding model can produce code that is just convincing enough to pass review even though it&#x27;s actually totally wrong. I don&#x27;t know whether this is a specific issue here.

jump in chart form <a href="https:&#x2F;&#x2F;x.com&#x2F;swyx&#x2F;status&#x2F;2064414823748886591&#x2F;photo&#x2F;1" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;swyx&#x2F;status&#x2F;2064414823748886591&#x2F;photo&#x2F;1</a>

Bummer! When can I finally and confidently get slopcode into Zig?

4.6 had functional but very poor quality code

I am shocked at the low scores from previous models. Maybe I just have low code standards but I&#x27;ve generally been vibe coding since 4.6

On the new FrontierCode [1] benchmark (ie graded from an OSS maintainer&#x27;s perspective of &quot;would I merge this code?&quot;)- Opus 4.7 xhigh: 5.2%- Opus 4.8 xhigh: 13.4%- Fable 5 xhigh: 29.3%Seems like a huge jump.[1] <a href="https:&#x2F;&#x2F;cognition.ai&#x2F;blog&#x2F;frontier-code" rel="nofollow">https:&#x2F;&#x2F;cognition.ai&#x2F;blog&#x2F;frontier-code</a>

Agree the per-task capability hasn&#x27;t been the blocker for a while. But on the autonomous-loop question — in my experience that&#x27;s not gated by how good the model is on any single step. What kills the loop is it slowly losing the constraints from earlier in the run and walking back decisions you&#x27;d already settled.

An 11% jump over opus 4.8 and a 22% jump over gpt 5.5 on Agentic Coding Benchmarks is certainly impressive.Obviously still need to verify it for myself to see if it&#x27;s truely a leap.But am I the only one wondering, &quot;What can I do today that I couldnt do yesterday?&quot;Previously I would think &quot;Oh I wonder if I can finally get it to do X now?&quot;However now I feel like yesterdays models were more that capable to handle nearly any engineering task I paired with it on.Maybe this is the final leap where I can comfortable set up an autonomous coding loop? Maybe.

Everything I&#x27;ve heard from people who have subscriptions is that they blow through their daily token quota sometimes in a matter of minutes, there&#x27;s rate limiting, etc. They spend a lot of time just waiting to be able to use it. And they&#x27;re paying through the nose for the privilege.It&#x27;s all a scam.

&gt; We expect demand for Fable 5 to be very high, and difficult to predict. On the Claude API and consumption-based Enterprise plans, Fable 5 is fully available from today. For subscription plans, we’d rather give access sooner than later, so we’re rolling out more conservatively, in stages:&gt; - From today through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost.
&gt; - On June 23, we’ll remove Fable 5 from those plans. Using it after that will require usage credits. If capacity allows, we’ll extend the included window.
&gt; - After this point—when sufficient capacity allows us to do so—we aim to restore Fable 5 as a standard part of subscription plans. We intend to do this as quickly as we can.I really wonder what their compute layout is for this. My guess from my understanding is that they know how to restrict during peak times and are willing to do this. Meaning we expect not the most fast responses and they can delay the inference to not have the service be down. Then, if that delay time is too annoying for token payers, they&#x27;re saying they should be allowed to remove cost by taking away the subscription users.

We, globally, can stop it. It has worked (so far) for nuclear disarmament, and could work for training large models. I know that policing the usage of computer clusters is not a popular opinion in technical forums, but something has to be done.Specially when talking about potential superintelligences. And if people think that&#x27;s impossible, remember that current models would have been considered science fiction just a few years ago.

It&#x27;s the &quot;If we don&#x27;t, someone else will&quot; effect. So long as there are competitive markets and competition between nation-states, a single player cannot unilaterally defect from the race, no matter how dangerous it is. Half the comments on HN lately are &quot;wtf Claude is so dumb compared to Codex; I&#x27;m switching&quot;--  nobody can slow down while those exist.

Curious what your idea would be here for a truly good actor in this space; no AI development?

It&#x27;s a five horse race between Alphabet, Meta, xAI, OpenAI, and Anthropic.Alphabet dropped &quot;don&#x27;t be evil&quot;; Meta&#x27;s CEO called their own users &quot;dumb fucks&quot; for trusting him and also clearly thinks &quot;super-intelligence&quot; is just a buzzword given how he tries to sell it; xAI&#x27;s model called itself &quot;Mecha Hitler&quot;; and OpenAI&#x27;s CEO was temporarily fired by the board for a lack of candor.It&#x27;s very easy to be &quot;the good guys&quot; with this competition.

Even if they are... road to hell and all that

The marketing has really, really worked for so many developers that will proudly and unironically proclaim that Anthropic are the &#x27;Good Guys&#x27;.

Please define what &quot;predicting the next token&quot; means. The next token according to what probability distribution? Couldn&#x27;t every process that produces text (including humans writing) be modeled as predicting the next token according to some distribution?

It doesn&#x27;t know. It&#x27;s not willing. It&#x27;s not thinking. It is predicting the next token.

No kidding. If my LLM issues commands to an agent to delete files I want to keep, that&#x27;s not &quot;intent&quot; or the model somehow become evil - it&#x27;s just a bad model that&#x27;s not doing what I want.But, for marketing purposes, it&#x27;s quite effective to portray your model as having some cosmic struggle between good and evil in itself.

This is all marketing, you don&#x27;t have to believe everything a company is saying about themselves, and you shouldn&#x27;t.Although, I could see Anthropic making a model purposely dangerous so there are bad outcomes and they can use that to their advantage for regulatory moats, and or in general make people think its more &quot;alive&quot; than it is. For some reason many people associate dangerous actions taken by llms with intent.

As much as I agree there&#x27;s a risk, we should still appreciate the fact it&#x27;s being disclosed upfront.

<pre><code> [Mythos 5] does sometimes still engage in reckless
 or destructive actions in service of a user’s goals,
 and our interpretability analyses indicate that it
 is aware that these actions are transgressive while
 it engages in them. As with Opus 4.8, rates of
 evaluation awareness and reasoning about being graded
 are significant, and not always verbalized; we
 introduce new and more detailed measurements of the
 nature of this awareness. The reasoning text from
 Mythos 5 is somewhat denser and more difficult to
 interpret than that of prior models, containing
 more jargon and difficult language.
</code></pre>
So, it (often) knows when it&#x27;s being tested while hiding that fact, is willing to break rules, is great at hacking, and it&#x27;s getting harder to understand what it&#x27;s thinking.Humanity has plenty of catastrophic risks to deal with already, I wish my field was not working hard to add a new one.

I swear I read a joke that &quot;what if we named chatgpt 5.5 Fable. Could we hype it as much as mythos?&quot; Last week!

I think Mythos is rumored to be ~10T parameters, so in this case I think the answer is yes, although I&#x27;m sure MoE, looped models, etc play a role in the improvements as well.

I&#x27;m a bit out of the loop, but do we have some grasp on the size of these closed models? Is the trick still adding an order of magnitude to weights and training data or has something changed?

They just leased a massive spacex data centre.

I&#x27;ve been getting a lot of these messages today:API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited

No issues that I&#x27;ve seen so far. Seems to be holding up for now.

Opus will be gutted furthermore. &#x2F;s
I feel 4.8 is very slow in last 2 days

I&#x27;ll be amazed if they manage to keep their infra responsive over the next 2 weeks.

I was just saying last week: If Opus 4.8 max is as good as we get, and we plateau there, I think I&#x27;d be fine with it.For the stuff I&#x27;ve thrown at it, that configuration has done a really great job. Including 70+KLOC go proxy with extensive test suite, some retro games, and more.

Kind of how DeepSeek v4 dropped their pricing? I sense a shift which will hopefully bring lower and lower cost. Then again Qwen3.6 coding has been all I&#x27;ve needed for my projects and I&#x27;m perfectly fine with free.

This is the entire business model of all AI companies. It costs far more to run the datacenters and build more capacity than they could ever hope to make back at current pricing models. I&#x27;m looking forward to pricing to catch up with reality and the resulting chaos that ensues.

Didn&#x27;t they admit this at the time? Cost was one of the reasons they gave for not immediately making it public.

Seems to me this is more honest than the Mythos claims a while ago. too powerful to release publicly. Too expensive?

I was just thinking this reminds me of the scene in The Wire where Avon admits to D&#x27;Angelo that the new heroin is in fact just the old heroin with different baby powder cutting it.

From today through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost.
On June 23, we’ll remove Fable 5 from those plans. Using it after that will require usage credits. If capacity allows, we’ll extend the included window.
After this point—when sufficient capacity allows us to do so—we aim to restore Fable 5 as a standard part of subscription plans. We intend to do this as quickly as we can.This seems like the pharmaceutical method of get them hooked on the drug with free samples, then once they can&#x27;t live without it, raise the price. I&#x27;m not sure I want to start using Claude Fable on a max plan if it&#x27;s just going to go away on June 23rd.But maybe the more charitable reading is that they didn&#x27;t have to offer this model at all on those plans and they are giving the standard free trial.

Yeah, it is also known as Claude Mythos 5

How much and what kind of data do you need to throw at these models to get a good design interface?

Now someone post the link about how it’s impossible for humans to draw a bike from memory.

This is all we need, that moment the Pelican put the leg behind the frame, we are all doomed.

I also look for this reply because i like seeing the follow-up reply saying that this is not a benchmark anymore because labs have gotten it in their training data.that reply never failed to come it&#x27;s basically a meme at this point

This is the reply I look for in all the new model announcements. Its fun to tell people that I judge models based on pelicans.

Pelicans hate biking in the rain (as do I).

Probably none. They probably have much better targets to optimize for than an SVG pelican or even SVGs in general.

Not as much as Qwen, since apparently 3.6 35B surpassed Opus 4.7 <a href="https:&#x2F;&#x2F;x.com&#x2F;simonw&#x2F;status&#x2F;2044830134885306701" rel="nofollow">https:&#x2F;&#x2F;x.com&#x2F;simonw&#x2F;status&#x2F;2044830134885306701</a>

How much money do you think they spent fine-tuning on pelican SVG generation?

Looks like Fable constructed the &quot;max&quot; &quot;looking&quot; pelican of the previous model for the &quot;xhigh&quot; output token count of the previous model.

I&#x27;ve seen this reply to Simon&#x27;s benchmark for 2 years running now, and yet you still see improvements and objectively-bad results over time from new releases, even when I&#x27;m sure every frontier AI team has&#x2F;had a person at least partially dedicated to better bicycle-pelican SVG outputs. Alas.

I don&#x27;t know. Just looking at the bike frames (specifically the fact that the AI generated bikes have rather unsteerable front forks), it&#x27;s clear to me that frontier labs aren&#x27;t spending much time tuning models to make bikes look coherent, which I assume is an easier task than making a pelican riding a bike look coherent.

I just run my own benchmark for &quot;draw an SVG with $animal driving $vehicle&quot;. I won&#x27;t post my choice of animal and mode of transport, but there are plenty of uncommon combinations to choose from. So far it&#x27;s a fun and visually intuitive benchmark that does seem to correlate with model capabilities

Simon has addressed this on virtually every new model release. He also has unpublished alternate prompts. But the larger point is: this is a fun experiment, not a serious and objective benchmark.

It was a completely useless test even before the labs trained for it.

The way I see it the benefit of benchmark isn&#x27;t to take Simon&#x27;s results at face value. It&#x27;s a template for your own benchmarks that are easy to visually evaluate.

It&#x27;s evolved from a funny, unserious benchmark to a tradition. When a major new model is released, I now always check the HN thread for Simon&#x27;s Pelican post. I&#x27;ll be sad when I don&#x27;t find it.When it started, comparing the progress between models was mildly interesting but everyone (including Simon) acknowledges it certainly leaked into the training data long ago.

I&#x27;m beginning to wonder how much of a useful metric the pelican is because surely the frontier labs must be training their models on pelican-artistry because of how well known your test is now?

Pelican for Fable 5 on default settings is a clear improvement on Opus 4.8Fable 5 default: <a href="https:&#x2F;&#x2F;gist.github.com&#x2F;simonw&#x2F;036bee5a703e7ec84e34efa97443828d" rel="nofollow">https:&#x2F;&#x2F;gist.github.com&#x2F;simonw&#x2F;036bee5a703e7ec84e34efa974438...</a>Opus 4.8 (the &quot;max&quot; one is closest to Fable): <a href="https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;May&#x2F;28&#x2F;claude-opus-4-8&#x2F;#and-some-pelicans" rel="nofollow">https:&#x2F;&#x2F;simonwillison.net&#x2F;2026&#x2F;May&#x2F;28&#x2F;claude-opus-4-8&#x2F;#and-s...</a>Now here are the Fable pelicans for all five of the thinking effort levels - low, medium, high, xhigh, max: <a href="https:&#x2F;&#x2F;tools.simonwillison.net&#x2F;markdown-svg-renderer#url=https%3A%2F%2Fgist.github.com%2Fsimonw%2F94fde31c34a0400c1d29f57e6a708e6b" rel="nofollow">https:&#x2F;&#x2F;tools.simonwillison.net&#x2F;markdown-svg-renderer#url=ht...</a>Low used 25 input, 1,929 output - 9.67 cents: <a href="https:&#x2F;&#x2F;www.llm-prices.com&#x2F;#it=25&amp;ot=1929&amp;sel=claude-fable-5" rel="nofollow">https:&#x2F;&#x2F;www.llm-prices.com&#x2F;#it=25&amp;ot=1929&amp;sel=claude-fable-5</a>Max used 25 input, 14,430 output - 72.175 cents! <a href="https:&#x2F;&#x2F;www.llm-prices.com&#x2F;#it=25&amp;ot=14430&amp;sel=claude-fable-5" rel="nofollow">https:&#x2F;&#x2F;www.llm-prices.com&#x2F;#it=25&amp;ot=14430&amp;sel=claude-fable-...</a>

System Card [pdf]: <a href="https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf" rel="nofollow">https:&#x2F;&#x2F;www-cdn.anthropic.com&#x2F;d00db56fa754a1b115b6dd7cb2e3c3...</a>

Claude Fable 5