Name: Opus 5.5 for Work
Uploaded: 2026-09-25T18:31:32.452Z
Duration: 52 min 36 s
Description: Opus 5.5 for Work

Transcript for "Opus 5.5 for Work": Alright. Hello, everyone, and welcome to our webinar on using Opus five five for work. We are so excited that you are all here. It's definitely a full, room today. So welcome, everyone. And after that intro video, I have the honor of saying introducing Claude Opus five five, our most capable Opus model. It's faster and costs 40% less to run. And we are so excited to talk about how you can use it for your work today. Just a few housekeeping things before we dig in. Importantly, a recording of this session will be available. It'll be sent by email within twenty four hours. So if you have others to share this with, you can send the link and they will be able to see it too. We will have Q and A turned on throughout the webinar and we'll take a few to answer at the end. You can submit your questions. We have a couple of teammates who will be answering some live and just go to the Q and A tab to access that. And then finally, give us feedback. We'll have a survey at the end, and we love hearing from you and hearing what you want us to improve or what you want to see next. So be vocal and, submit the survey at the end. Now I'll introduce the team that will be, presenting with me. I have my amazing teammate Nathan Wu, who's on applied AI. Welcome on stage. And also Nick Lin, who's on our product team and has built many of the things that you will see today. So welcome all of you and you'll be hearing from the three of us. We'll be here for about an hour. We have some slides, but then the majority of our time today will actually be spent in the in the product, working on a project live using Opus five five and showing you where it shines. This webinar is focused on non developer, use cases. We will have more developer content available, in the coming weeks, but today's focus is Opus five five for work. And our demos will be shown in co work, as the main product surface. So let's talk about what changed from Opus five. So the main thing that's changed are, there's three parts. We have decreased the price of Opus five five compared to Opus five. We've made it run faster, so you can get more work done with fewer tokens. And finally, we've made the communication better. So you should enjoy talking with Opus five five more than other Opus models in the past. And we think that it's, shorter and clearer about what to do next. So we'll go into a little bit more detail on each of these parts. Opus five five costs less per token than Opus five and also uses fewer tokens per task. So in our evaluations, this nets out to about a 40% drop in costs. The levers shown on the right hand side are what, the pricing levers that we've moved to kind of pass this, efficiency on to you and our users. We've decreased input tokens 20%, output tokens 20%, and importantly, cash reads, have dropped 60% in price. Cash reads are increasingly important as you do more agentic work in Claude because every time you type a new answer, Claude has to read and reload the content that came before it. And that can be expensive if not, if it's, pricey to have cash rates. So we've dropped that the most as a result. Finally, we have faster output as well. You may notice Opus five five gets to an answer more quickly and also takes fewer tokens to get there. So we've seen this, that Opus five five generates output more than 30% faster than Opus five. And that means you can finish tasks sooner and not be waiting for for each reply. Finally, and this is what I'm most personally excited about is better communication. We're seeing our Opus five five communicates more naturally. The writing is clearer and easier to follow. And one of the more noticeable things is that it puts the most important information upfront. So we can see this on the example on the right hand side, particularly in this problem section, where it's it's pretty hard to follow, the OPUS five answer on why these orders are timing out again. It's a little bit, so it's not that easy to read, whereas OPUS five five has much more skimmable responses. So it's very clear on the right hand side that these order exports have failed twice this month and what we're going to do about it in the immediate term and the long term. So it's just writing a little bit more closely to how, we think. So although we've made a lot of improvements, and Opus five five feels like a different model, some things haven't changed in our model development, and that is our our prioritization of safety and alignment. So, a few things we're excited about here. Prompt injection protection has been built into Opus five five, and it's actually the hardest model to hijack that we've ever tested. So hidden instructions within prompts only succeeded 1% of the time within 15 attempts, and that's tied with Fable five one. We also had Opus five five independently evaluated by four reviewers, before launch. And we have fable level cyber and bio safeguards built into this Opus model. When a request is blocked because of the topic or the type of prompts that you, asked, the work will continue on an earlier model instead of ending in a refusal, which we think will kind of keep you in the flow with Opus 5.5 as you work. And finally, Opus 5.5 has achieved the best behavioral alignment score that we've seen. We anthropic runs an automated behavioral audit on every model. It's the most comprehensive alignment test that we run and Opus five five had an exceptional score. So with all of this in mind, our main call to action to you all today is try using Opus five five as your daily driver at work. We think it's a really pleasant model to use and an exceptionally powerful and intelligent model as well. And I'm now gonna pass it over to Nick, who will explain some of the ways that, Opus five five can be used in your work and where it shines. Alright. Thanks, Amelia. Now Opus Pfeiffer, as we mentioned, is really built for daily driver of enterprise work. A couple of things where it's really far superior compared to our past model generations, number one is alpha quality. It follows instructions precisely and stays in scope. So outputs like spreadsheets, slides, and documents need very little editing. It's also fantastic at data analysis. So especially for finance and data use cases, it reads dense filings, models, and charts very accurately and is able to carry context across a whole project or a whole deal. And more on finance in just a bit. And Opus five five is also our best Opus for vision. So it is very reliable for document extraction and visual analysis across charts, screenshots, and diagrams. Next slide, please. And Opus five five is really built for, as I mentioned, the data and financial analysis. So as deeper reasoning and precision for financial work, it scored 65% on big finance bench, and that really means that it is able to parse through long documents and spreadsheets behind real data analysis. And it is able to do so with much fewer tokens and way ahead on the cost per task across our curve. So the four things that really matter, it is able to read through dense data sources and materials across filings, models, and charts. It is able to keep the whole picture, full context, really throughout a full deal. It shows its work. It verifies its work. So clear summaries of what it found and how it found it, including compliance sensitive work, and it produces outputs that are ready to ship. Next slide, please. Alright. Now let's see this in action for an investment banking use case. Play, please. How many So now we're watching actually a video of Opus five five going head to head and creating a merger model and a football field output compared to Opus five. As you can see, Claude works where you work, Excel and PowerPoint. It's really built to check its work with even more detail. And Opus five five on the right, as you can see, has already finished a lot of this work way faster than Opus five with significant improvements in app quality and comprehensiveness as well. We're seeing that Opus five five is about 30% faster, and on this task specifically, almost 90% cheaper than Opus five. So we're really excited about Opus five five's ability to both improve the quality and speed of achieving enterprise tasks while maintaining even higher standards when it comes to quality. I'll pass it back to Amelia to talk a little bit more about how to actually get the most out of Opus five five. Thanks, Nick. And isn't seeing the models head to head just so cool. We can see how much cheaper it is and how much faster Opus five five runs. We think this really tees you up to be able to bring more ambitious work to Claude Opus five five compared to what you may have tried with Opus five. With with every model, there's a chance to improve something that we call, AI fluency at anthropic. And part of being fluent in AI is really retesting what is possible with net new models as they come out and what gels well with their specific capabilities. So this slide gives some ideas of things to try. One is, asking for the finished file. So Opus five five is really good at doing end to end work and giving you back, like, a full spreadsheet or a document instead of giving you back, like, the outline and then asking more prompts to get to the finished product. We think you can go end to end in one prompt more often than not. Next, we also find that Opus five five is really good at, nuance checks of long documents. So, giving it a long plan or a report or a long deck and asking it to find mistakes or discrepancies. Opus five five is trained to pay even more attention to detail than prior Opus models. And so, we found that it's able to catch, really small and large mistakes, including in one of our tests, it caught a chart, in a report that didn't match the numbers in the corresponding deck. We were able to go back and fix that. Finally, we think that Opus five five is also better at longer research tasks with open ended research questions. It's able to pull sources together, work through long documents, and turn what it finds into memos or reports or a deck. And part of the benefit there is that that's where Opus five five's communication really shines because what it's handing you back will be well thought through and, easy to read and understand as well. And then finally, this is where I've been personally pushing Opus five five the most is asking for visual creativity. Because of Opus five five's vision strengths, it's better at generating creative outputs and visual design. So try using it for prototyping UI, creating websites, or illustrating concepts. And then it wouldn't be an Opus five five for work webinar if we didn't give you some tips for using this model. There are a couple of behavior changes and kind of prompt updates you may wanna consider implementing as you get to work here. One of those is not telling, the model to think hard. You can take this verbiage out of your prompts. Opus five five always thinks before it replies and decides how much thinking, to apply based on the effort level that you set. And we think that or we found that dropping think carefully from instructions and prompts makes the reply start sooner. You can default, working with Opus five five to medium efforts. We found that's like the best calibrated for cost and performance with this model. Another, prompting behavior to to update is telling Claude what done looks like, in your first prompt. So you can actually prompt Opus five five with your goal or desired output in a single message instead of going turn by turn, progressively giving it specific instructions. We find that opus five five is much better at continuing work over long multi part, work, and it does that better than opus five. And finally, another point on design and vision. If you are having, Claude Opus five five, work on a design for you, name the styles that you don't want to see. With no direction, this model will fall back to a few default styles, which we think look really good. But avoiding saying, like, avoiding a generic look will just swap the defaults that Opus five five uses and giving it a list of specific patterns or anti patterns, that you want applied will work better than, an open ended, you know, avoid generic, design. And then finally, in terms of getting the most out of Opus five five's vision capabilities, we think that you should be sharing charts or screenshots of charts and diagrams, themselves. You don't need to copy or retype, specific data metrics into the model because this model is really, really good at reading those, itself and reading them accurately. And it doesn't need extra steps, in order to know to, incorporate those vision capabilities. So we'll show some of these tips off in our demos. But one of the things I wanted to mention is that we also have new and updated artifact types available in Claude. So we have docs, slides, and design directly available within co work. And these surfaces in our product tied with Opus five five's capabilities, makes for a really exciting new ways of working. And we think that you'll be able to create more impressive work outputs when you use these. They were shift default off in enterprises. So admins can go into artifacts, under org settings and turn them on to specific groups, within enterprise plans. But I wanted to introduce these before we show them off in our in our demos. And now, Nathan and I are gonna take our presenter hats on and put on our personas for for the demos we're gonna show off. We'll be walking you guys through a single project, but have three, modular demos within. Just so that we go in with, same premise in mind, where we're going to be discussing a retail company with declining growth. I'm going to be playing the role of Nathan's manager and together we're going to investigate why is growth declining and create a presentation for the board. We'll walk through some data analysis, show off multiplayer, and then I'll show head to head what looking at slides in with made with Opus five versus Opus five five looks like. So with that, Nathan, I'll hand the screen sharing over to you and we will transition to the product itself. Awesome. Thanks, Amelia. And, hey, everyone. Thank you all for the time today. I'm Nathan. I'm a member of our applied AI team here at Anthropic. And today, I'm super excited to be able to showcase to you all some of the exciting ways we can use Opus 5.5 combined with some of the new features in Cloud Cowork to unlock new and creative ways for how you can do your best work. So here I have pulled up a session of Cloud Cowork. As Amelia mentioned, the setup is Amelia is my boss. She has some questions about why revenue growth has slowed over the past year. She has some ideas of why that might be the case, and she'd like me to look at it further. This is precisely the type of open ended data oriented task that Opus 5.5 is really good at. Opus 5.5 is really good at noticing data discrepancies, discovering patterns within data, and generally just working with you as a smart collaborator on your hardest, most open ended tasks. So before we dive too deep into the demo, let me orient everyone, especially those who may not be as familiar with co work and what we're looking at here. In order for Claude to do its best work, we need to give Claude some information that it can use to answer the task at hand. This is called context, and here on the right hand side of my screen, you can see I've already pre uploaded a couple of different context, files, a k a data sources for Claude to look at. You can see we have a couple different data types. We have some Excel CSVs naming our stores and their sales. We also have some PDFs from the latest, leadership off-site. We also have a bunch of other different data sources as well, and you can see it spans a wide variety of structured and unstructured data sources. So in this case, I uploaded the context directly into Claude, but Claude works best wherever it is that you work. And another way to upload context into a, co work session is via what we call connectors. Connectors are essentially integrations to your existing business systems and data stores so that Claude can access this information just as you, a human, would. So, for example, let's say you already work in Google Drive and all your files and the data that Claude needs for a task is within Google Drive or it could be Microsoft SharePoint or a database like Google BigQuery. As long as you have this connector enabled, Claude will be able to search those data stores, just as you would. Chances are for any common business system or data source that you work with today, we already have a prebuilt connector available for it. So as long as this is toggled on, you should be able to get the most out of cloud wherever it is that you work. And please ask your admin if you would like, additional access to, other sorts of connectors. So here, before we get started, just a few other quick pieces. On our model selector here, I've selected Opus 5.5, as the model for our task at hand. I'm using the medium effort level, which Amelia discussed as a good default for your daily driver work. And let me start ahead with the prompt here. So as Claude is working, I'll just read sort of the task that we gave it. We mentioned, hey. Gross slowed a lot this year. Leadership has a couple theories about why. They're in Amelia's note. I gave Claude a big data dump of everything that I have in an uploaded folder. Please test each of the four theories against the data. Tell me which ones hold up. A couple notes about the prompt here. One, you may notice that we don't have specific instructions like think really hard or make no mistakes, and that's because Opus five point five is already trained to do the best work without needing these sort of additional instructions. The other thing that I would call out is that if you look at the prompt itself, I wasn't very prescriptive about how Claude should go about solving the task. Instead, I gave Claude the goal of what I'm looking for and how I would say that it's done, AKA please test the four theories against the data and tell me which ones hold up. Instead of telling it step by step, hey. For this individual hypothesis, you may need to look in this Excel file or this PDF, to get the answer. And this is in line with our general theme about bringing your most ambitious work to Claude, giving it goals, and working with it as you would a smart collaborator versus prescriptive subtasks step by step. So it looks like Claude has, given its answer here. It's gone through each of the four theories. It looks like it's pulled some additional data. One of the great things about working with Opus five point five is given that it's such a smart collaborator, we can interact with it as we would a colleague and ask it sort of follow-up questions. So let's say, hey, Claude. I think the issue might be related to our discounts and promotions. Can you please look into that? And because of some of the improvements in the communication style that Amelia walked us through earlier in the presentation, we think Claude is much more pleasant to interact with. It speaks in plain English more concisely, and we think this lends itself really well to this sort of iterative back and forth workflow that you're seeing here. Another point I would call out is a lot of the work that you're seeing is being done in an agentic manner with a very, big goal and a long kind of horizon to accomplish this task with a bunch of different steps. And because we've aligned Opus's Opus five point five's pricing to better reflect this kind of work behavior, you don't have to worry about running out of usage or overshooting on costs because you've added too much context or, taken too many terms with the model. So I'm gonna fast forward a bit here, and let's say we're pretty satisfied with what Claude has produced, and we'd like to actually produce a report to share with Amelia. Cloud works wherever you work. So in some cases, maybe you would like to work in Excel, maybe you like to work in PowerPoint, maybe you like to work in Docs. Cloud can already produce that natively for you, today. For today's demonstration, I wanna showcase a new surface that we have in CloudCore called Artifacts, which you can see here, and specifically the document type artifact, which Amelia briefly mentioned in our opening presentation. Effectively, this is just a text editor, that you can use just as you would any other text editor. But in this case, we do not need to access artifacts from the sidebar here, and instead, we can directly access it from the Claude cowork session where we are working. So here, let's say we're satisfied now and, hey, Claude. Please make this into a report for me to share with Amelia. Please use our house style. And as cloud begins to work, you should see that it's starting to read the different files that it needs to, do its best work. And in a second here, we should see a, pop up for the docs, tab, which will showcase, the new artifacts directly in CoWork. And while that's loading, you can see that, Claude is really good at, reading through different files, taking different steps. One of the improved things about Claude CoWork in the communication style is that it's more verbose and explaining, why it's doing certain things. One other thing, as it's pulling up Cloud Docs, it looks like in order to use this feature, make sure on the connectors tab, you have Cloud Docs enabled. And, hopefully, in a second, we will see it, come up. Alright. Love the live demos here, but maybe we'll jump ahead to the actual output. And you can see this is an example of what the Cloud doc output would look like. You can see that it's produced a couple of, bullet points in line with the explanation that it gave us earlier. And one of the things we really wanna emphasize about Cloud Docs is that it's really a canvas for you to do your best work. And so what that means is a Cloud docs supports a lot of things natively besides text, such as charts, interactive visualizations, animations, etcetera. And so just to give folks a sense of what, is possible in Cloud docs, here, we can see Claude has made a table of contents, made some statistics. It's also started putting together some charts and visualizations based on the data analysis that we did. And as I scroll through, you can see that Claude has made an interactive visualization animating for each individual hypothesis whether or not it was supported by the data. It looks like I earlier asked about discounts and promotions, in which case, Claude determined that is not, the case and produced a chart, for, backing up its claim that that did not impact growth. It also noticed that, hey. For one of the hypotheses, we can't fully rule it out, but, it's not present in the data. And at the end, produces this infographic showing the different hypotheses and how they weren't supported by the data. However, Claude independently assessed that there's a new possible explanation for, what might be happening here. And in this case, Claude is proposing that the reason sales, have declined is because the new stores that we open are actually too close to our existing stores. So it produced this nice animation for me, which is that, hey. Look, at our existing stores in green. Once we add a new stores in black, you can see that the growth rate declined. We actually look at the stores, listed, by city. The pattern's a little unclear. But if we filter the stores instead by their proximity to a new store, then the pattern becomes more clear. Claude also builds this nice regression analysis showing the relationship between distance, of a a new store and an existing store and the effect on sales. It also took an analysis of our existing, old stores and the declining sales and showing that a lot of it went to the new stores that we opened. And at the end, it produces this nice chart illustrating the different, impacts on growth from the analysis we just, showed, and you can see that older stores, losing sales to newer stores is the primary decliner, and driver of the the fewer sales, this year. So this is one example of cloud producing some nice charts and visualizations. Additionally, cloud can also produce things that are interactive with your audience natively within cloud docs. So let's say, for example, for my executive audience, I want to showcase, a slider here that says, hey. How does our sales, change if we change the distance, from a a new store to an existing store? You can play around with this, create other widgets, for folks to, to to explore the data further. And one of the other cool things about Cloud Docs that I wanna showcase is we heard a lot of your feedback about Cloud cowork feeling like a single player story. And so I'm excited to share that Cloud Docs actually now supports natively, multiplayer behavior. And so let me add Amelia into this project here. Yay. Yeah. Perfect. Perfect. Thank you for. sharing that, Nathan. Are you able to access the project? I am currently looking in my artifacts tab, and I see the why did growth fall from 13 to eight. And now you can see my icon in the top right. So if I highlight this, you're you're able to see that I'm also in the doc and we can start collaborating together. How should we how should we improve this? Yeah. This is really cool. Amelia, do you have any comments or observations about the document, anything you'd like changed, the wording, etcetera? Yeah. I think, you know, one of the things I wanna know is, like, what's the source for this number? And I'll kinda go through and just leave leave some comments here. But I'm very excited about the way that these charts are, like, super interactive and grabbing my attention and making it clear that, you know, a lot of my hypotheses were actually incorrect and Claude came up with a new one on its own. So that's awesome. Great. And so not just, interacting with humans like you saw there with your your colleagues, we can actually also interact with Claude as well. So let's say in, one of the, charts here, I don't like how it looks. I can actually tell Claude, hey, Claude. Can you please change the squares below into smiley faces? And you can see by tagging Claude, Claude automatically starts working on the task at hand, and we think this represents a new and exciting form factor for how, we can work both together with other humans, but also with all of our agents in the same workspaces, within one co work session. The cool thing is not just me having the ability to interact with Claude, but my colleagues are also able to interact with Claude as well. Amelia, do you wanna give a shot at trying to ask Claude to do something? Yeah. So I have this, like, sales fell at older stores highlighted, and I wanted to ask, Claude, like, how many older stores are there. But I am only a viewer. So could you upgrade my permissions? I'm gonna ask to edit so that I can also add Claude within this doc. Awesome. It looks like Amelia's requested access, and I will approve her as an editor. And once she is approved, are you able to now tag Claude as well? Yes. We're all we're all set. So this is awesome. Let's ask Claude some questions about the data and then we can move to get this board ready. And. we'll start working on it. Yes. So this is one example of how we can work together with Claude, and people in a a new and interactive format. We hope you'll try out artifacts and hopefully gives, you a sense for the new and creative ways you can use Claude today. I will say that, for a lot of folks and a lot of our customers, even though docs is a really new and exciting form factor for us, we recognize that you may have worked in other places, or other file types that you prefer to work in today, which is all great, and we will cover that, in a second with Amelia. But I just wanna showcase that, hey, once we're done with the docs here, if we want to export this into Word, into PDF, into Google Docs, into Notion, we support working wherever, it is that you work today, and so we really want Claude to be present where it is that you do your best work. It looks like perfect timing. Claude has updated our chart here to smiley faces, which is really awesome, and sad faces, as the the growth goes down. So thank you all for for taking the time to, listen to me walk through docs. And now I'll turn it back over to Amelia to walk through, the next step of our demo. Yes. Thank you. So let me share my screen once you pull down yours. And let's see. So now you can see my Claude instance. And, you know, I asked Claude to shorten one of these lines and it's working on it live. What we found is that working in docs together, like, really has brought to life some of our meetings and discussions around around data, at anthropic. And we can all be kind of storytellers with some of the data analysis Claude pulls together. A lot of, like, enterprise work is translating data for different audiences. So Nathan passed on the doc to me, and now it's on on me to own kind of getting this into a board ready presentation. Within any doc, you can always go to chat and then chat directly, with it or pick up work, just having in kind of a split view. So what I wanna ask, and I'm gonna switch from Opus five to Opus 5.5 here. What I wanna ask Claude to do is, hey. Can you make this doc into high level slides on the takeaways for our board? And then I have a pretty strong opinion that I want it very clear, like, which hypotheses, were correct and which ones were wrong. So make sure there's a summary on which hypotheses were disproven and why. And then I'll hit send on that and Claude will be able to turn this stock into slides for me. Slides are one of the artifact types that you can, click into and create. One entry point is going from a doc to slides, but you can always click into these and just start a new presentation fresh, as well. So we want you to be able to kind of think through your work or start, net new work on its own. So here we can see Claude is starting to work, creating slides in the sidebar. We're gonna kind of let that run-in the background. Sometimes slides can take a little bit, but they are much faster because of, Opus five five's, abilities and speed. So what I wanna show is actually a head to head that highlights some of the visual differences between an Opus five generated output versus an Opus five five output. So if we set aside the retail, story for a second, I wanna go into, one of the other things I was working on, which is around this customer summit that we had. So I started, with with, chat in Opus five about, set of feedback that we got in a Google Sheet, from a from a customer summit that we held. And it's a mix of good and bad feedback, but my, Claude instance is connected to Google Drive, so it knows to go look at that sheet and pull in the information from there. It's creating slides that kind of show the responses. And what I wanted to show is how I took its original output, which was like a little bit corporate. We can take a look at it here. It's a little bit plain. It's just a debrief, and it's kind of giving, like, the overarching story, but it's not it's not particularly well spaced. These are different fonts, it looks like, or font sizes. And it just doesn't quite have, like, the crispness that I would want in a presentation. It's, like, impressive that it came up with this, but I'm finding it's, you know, text heavy and has some of the hallmarks of, like, Opus's communication, patterns. What I did, with Claude and one of the prompts that I kind of suggest that, is worthwhile to give a try is to ask Claude to, like, see what this would look like if you didn't create such corporate slides and instead made something creative that told the story. And so I did that as well, and we can see that this is starting to get the hang of, like, this the it's now talking about these two summits, but it's also still a little bit wordy. And some of these things are not well spaced when using Opus five. Let's switch over to seeing, what five five came up with. When we look at these slides, the original version is like pretty clean, pretty crisp. I like the font selection. It's well spaced. In particular, what stands out are actually these slide titles. This is very easy to understand, exactly what happened at this conference. Newer users rated us really highly. Tenured accounts did not. And it's showing the ratings for different presenters. It's still a little bit, like, flat. And so when I had it come up with a, you know, a little bit spicier version, making them more creative and focusing on storytelling, it came up with this version, which is, we threw one summit, but our customers went to two. It's leading with the thesis. Some customers rated us a 10. Another set thought it was a waste of time. And importantly, it comes up with this very nice distribution that that tells a story really well. So this is all to say I'm much happier with the Opus five five slides. It's there's, like, more variation in the slide types. The headers are clear, and it's not trying to cram too much information into each slide. I think if I had to sum summarize, like, the difference between five and five five for slide creation, it really comes down to, like, it seems like Opus is looking Opus five five is looking at, like, the full story that I need to tell. Whereas, Opus five attacks each slide, like, individually and doesn't lead to such a, holistic, like, good rhythm, high empathy, like, what would my audience wanna see in these slides kind of output. So while we were going through that, we should be able to see, mock of a board deck on why our growth at that retail company, was slowing. So as a reminder, I took Nathan's doc and had it convert it into a board readout, and I've never seen these before. So let's go through them together and see, what we like or dislike about, these slides. So within here, this is clear growth halved, and our new stores are the main reason. Love that it's putting the point upfront. It's doing these four explanations. Now I feel like this could have been a little bit more, visual here. So I may just, like, drop a comment on the whole slide or just the table and be like, can you make this into columns instead of a boring table? And Claude will get to work on that in the sidebar. And then I'll leave the comment functionality and progress, and we'll see it replicated the chart really nicely and has, like, well balanced colors and composition. So I'm pretty happy with these as a first a first draft. The copy needs a lot less tweaking, and I'd feel comfortable kind of making some edits and sending them off to the board. I can also share these back, with Nathan. So he is able to, Nathan. So he's able to also get in and see if it's making the key points. It initially adds him as a commenter, but I'll also upgrade him to an editor right away. And I can also copy this link to share it as well. Now, if you haven't gleaned, the main message we're trying to say about working with Opus five five and the direction of anthropics products, it's that we're really trying to make our products and models excel at the work that you do in your everyday job. We think that Opus five five is like speed, the price where you can actually have it take on these tasks and know that your tokens are going to go further than they did with Opus five, or some of our other models. And the fact that it's much better at vision and communication will make it something that you can use as your your daily driver. And you'll see Claude get more and more collaborative as a surface. Now one of the hallmarks of of enterprise work is that we know you all have many different tools that your teammates work in. So if there are, other places that your team works, you can always send this to different apps or different formats. Bring it into PowerPoint and continue work there or expert in export and share as a PDF. So we want, whether your team is in cloud or not, we want your work to be able to shine across across different surfaces. And with that, we've now gone through some of Opus's capabilities and how they show up in in co work, and seen how Claude is really good at data analysis, spinning up docs and slides. And then I think we can move into some Q and A. We're gonna take a minute and kind of find where the questions have come from and what are the most upvoted ones. And we'll take a stab at answering some of them live together. I'll pick off one. Let's see. So one of the questions was around, like, whether or not these tasks can be done in cloud chat or if they're only available in co work, you can, create outputs also directly from chat as well. But going back to one of the key pricing levers we pulled for, making Opus five five, more cost efficient to work with in co work, we decreased the price of cash reads. And cash reads are one of the reasons that co work as a more agentic product surface was more expensive than chat. And so we'd love for you to retry kind of trying some of the tasks that you've done in co work, with Opus five five and seeing if you're able to have your tokens go go further in the surface than before. Nathan, any questions you you see that are rising to the top? Yes. I've seen a couple variations of questions that are, saying, hey. I'm not able to see artifacts or, hey. How how do I get access to this? If you are on an enterprise plan, your administrator needs to turn on artifacts, and so I would definitely recommend chatting with them and making sure that it's enabled in your Cloud Co work instance, for your group or workspace. And your administrator can also turn on, these surfaces for a subset of users, to kind of test, see how they work, create some templates for the rest of the team to use, rather than it's not just an org level toggle that will is either on or off for everyone. Let's see. In terms of, using OPUS five on high effort versus five five on medium, every model that we put out, we run a lot of evaluations across different product surfaces, to hone in on the, like, recommended default for balancing both cost efficiency and capability. And so that's really the difference between Opus five on high effort versus Opus five five on medium. We're finding that medium effort goes an extremely long way with Opus five five, given its intelligence, and capabilities, as well as its token efficiency. Importantly, one of the things we've seen with Opus five five is that running it on, like, max or extra max, won't always, like, get you a much, much better output. We think that medium is really, like, the sweet spot as a as a default. I see another question in the chat about, hey. How can we point, you know, Claude docs or Claude presentations to use our brand guidelines, to create graphics or presentations in, in line with our existing style. So in the example that you saw, what we actually did was be, if you have already instructions labeled out somewhere, Opus 5.5 is very good at following those instructions. It may take few iterations to get it precisely, the way that you like it, but that is definitely one way to do it, and we actually did that live in this particular demo to follow Anthropic's brand guidelines. The other way that you can do it instead of, you know, giving it instructions verbatim is if you recall in the demo, we showcase our connectors features. And if you have existing examples of graphics or presentations that you really like or you think are gold standard, you can point Claude to them. Let's say they live in Google Drive and say, hey, Claude. Just for reference, like, our best brand guidelines or the best decks that we use today, live in this folder in Google Drive. Please make sure you reference them when you create your output. That is also a very effective way of getting Claude to adhere to your, brand style. Yeah. It's very it's very, powerful when you can kind of say, like, use anthropic style or as a Nathan's demo, like, the house style, and have it have it preloaded in, as a direction. I see a lot of questions around, like, when to use SONNET versus when to reach for OPUS. And we think that in this in this period where you have OPUS five five, released, and because of its, like, pricing and, token efficiency, we really do want to see if you can switch to OPUS five five as your daily driver, and see if you're able to get the outputs that you want, from from OPUS five five. We did announce that we're launching a model family, and so a new SONNET will be coming. And you can, use that as well as the default if the cost is still too high on Opus five five. But generally, we're finding that Opus five five is, exceptional, daily driver and, price performant one as well. Maybe one question here. How you describe the new design artifact versus the dedicated cloud design functionality? So with the caveat that all of these features are still in beta, but definitely, we encourage you to test them out today. The intention is for everything that you could do in the standalone cloud design app to be replicable within the, co work or chat, artifact, for design. So the goal is to make them one to one. Yeah. We fully ported over all of the existing functionalities from design and added quite a lot of interoperability across the document types. So that is one big benefit of making sure that the cloud app is just the one place that you can go to for all of your output needs. And there is a question of, whether we have combined chat and co work. On certain plans we have, we're still in the process of rolling this out. So some of you may see one unified tab, and some of you might still see the chat and co work as being two separate tabs. We're going to roll this out to everybody in the coming couple of weeks. But the very similar types of use cases will automatically get the co work style harness, as we call it. So you should be able to see your high quality outputs in any case. Yes. And importantly, the the outstanding, like conversion of enterprise is going to be happening. It won't be on by default. We we have a transition period for getting, the combined chat and co work experience within enterprises. One of these questions is also about, like, when you export slides to PowerPoint, are the content still, editable? And, yes, when you when you export a slide from CoWork into the PowerPoint surface, you'll see the components, render on the slide. And it should be pretty similar to to what you saw in the slides themselves within the the clogged surface. Let's see. Maybe one more one more question and and we can kind of wrap up for today. We're so excited for you all to to give OPUS five five a a try. And please share feedback with us, through any any channel. We're always watching and listening to to users about what they're achieving with Opus and, our cloud prod projects as well. I think some of these, questions are around, like, the importance of context, and how to bring context into co work or a different cloud surface. And this is probably, like, the single highest leverage thing that you can do to get, really, really good outputs relevant for your work out of our cloud services is have it connected to your enterprise tools and where your work lives otherwise so that cloud can go and reach for what it needs when it needs it. Nathan's demo showed tons of data dumps and reports and files and notes, and that's kind of what, any worker needs to do a really good job. So that connector piece or adding context into cloud, it's like probably one of the more salient takeaways from today. But I am so happy you all joined, and I'm so happy you're all excited about Opus five five. Thank you to Nick and Nathan for presenting with me and crushing it today. And please fill out the survey. We'll be sending across some, resources as well in the email that you get. And as always, the recording will be available for you to pass on to your teammates, or watch pieces of this again. Thank you all. Thanks, everyone. Thank you all.