我對比了GPT-6 Astra 和 Fable 5.1，結果出乎意料！

Over the past week, I have been testing both GPT-6 and Fable 5.1. I used various daily life scenarios for a side-by-side comparison to see which model is truly better suited for our daily work. For instance, I had them simultaneously help with performance reviews, act as external consultants, draft sales and video scripts, create slide decks, and utilize the computer-use function. For every scenario demonstrated today, I will provide a detailed analysis of which model performed better, how long each took, and the associated costs. So, if you want to know which model is best for which task, this video is definitely for you. Without further ado , let’s get started right away. First , before we dive into the demos, I want to clarify that all the test examples are designed based on real scenarios from our everyday life and work. I know the GPT-6 launch featured many cool 3D effects and demos for gaming and animation, but as some of our viewers noted, many of these scenes have little to do with our daily routines. Guided by the principle of helping you use AI effectively, our cases today won't involve building games or 3D effects; they are purely focused on improving work and life efficiency. Also, I should mention that for this test, I subscribed to three or four accounts for each model. Because token consumption was so fast, I spent several hundred dollars just on usage fees, so I sincerely hope this video provides real value to you. In addition , all the high-value prompts used in today's video are provided for free as usual; you can find the link in the description and the pinned comment. For the first scenario, let's have both models conduct our mid-year performance review. I believe every professional needs to undergo performance evaluations at the end or beginning of the year. Rather than digging through folders or emails to find evidence that you've met or exceeded your goals, simply keeping track of meeting records allows AI to summarize your mid-year review and save you a lot of time. For example, here is a complete set of instructions I gave it, asking it to analyze all our previous meeting records and automatically extract content relevant to the performance review. As you can see here, I have uploaded my performance review form and the minutes from every meeting. Of course, you could also just have the AI pull this information directly from your work folders. First, let’s look at the results from GPT-6. Following our requirements, it provided a summary evaluation of our mid-year performance, along with three reasons behind that rating. Next is the itemized performance assessment and corresponding review, supported by key evidence. As you can see, the level of detail is very high—it even specifies the exact date each of our documents was issued. It then highlights the five achievements that best reflect our performance, as well as tasks that haven't met targets yet and what needs to be done to reach them. Of course, we definitely need the AI to help us list the extra work we’ve done beyond our original job requirements, as that’s the only way to effectively ask your manager for a raise. And naturally, there will also be a mid-year performance self-assessment at the end. Now let’s look at the results from Fable 5.1, where we used the exact same prompts and uploaded the same files as we did with GPT-6. Similarly, Fable 5.1 also followed our instructions by including the mid-year performance summary and its three main reasons, the itemized performance review, indicators that didn't meet targets, work beyond the original job scope, and the mid-year self-assessment. It was also very detailed. And look, Fable 5.1 even helped us generate the self-assessment text that can be copied and pasted directly into our performance review. I found that compared to GPT-6, the results from Fable 5.1 are more organized, allowing me to understand the output in a short amount of time. However, GPT-6 does have its own strengths, specifically that it is superior to Fable 5.1 when it comes to detail processing and content richness. Finally, let's look at the time and cost spent by both models on this task: GPT-6 took 2 minutes and 56 seconds at a cost of $ 5.60, while Fable 5.1 finished in just 2 minutes and 30 seconds for a cost of $ 4.40. For the first task, I think both sides are actually evenly matched, as each model has its own strengths. So next time you have a performance review or need to summarize a meeting, I suggest that if you can, use both models: use Fable 5.1 to help you organize the overall framework, and use GPT-6 to help you fill in the details. The following case will definitely be of great help to everyone. I believe that at the end or beginning of every year, you all set goals for the new year, and at the end of each quarter, you look back at your progress, or if you are investing, you also hope to get professional advice. In all these scenarios, and any others you can think of, we can treat AI as our strategic consultant to provide us with practical and professional advice. For example, here we ask both models to help us set annual goals, and at the end of each quarter, help us review them and adjust our plans. Of course, here we can also let the AI access content from our knowledge base. For example, here I have it go into my Google Drive and Gmail to extract relevant important documents and emails for us. Thirdly, here I tell it to provide advice based on what it already knows about me, which is essentially leveraging the AI's memory storage. Here I have them first tell me what three things I should focus my energy on for the next 12 months. If it's a quarterly review, it helps me re-evaluate the goals I set and the current progress of my plans, and then provides corresponding adjustment strategies based on the actual progress made. Here, GPT has already conducted a line-by-line review for us, finally telling me the three things I should focus on for the next 12 months, and additionally, it tells us the three goals we should pursue for the remaining 90 days of the year. I think a particularly good feature of GPT-6 is that it allows us to schedule quarterly reviews, and this option is already enabled here, so you can adjust the notification frequency, and when the time comes, GPT will automatically send you a notification. Now let's look at the results from Fable 5.1. First, what I really like is that Fable 5.1 is exceptionally organized in its output. As you can see, it gives us the conclusion right away, and then explains the reasoning behind these three points, moving from the big picture down to the details. It tells us what to stop or reduce, what deserves more time and resources, and it even identifies the biggest bottleneck we're currently facing. Furthermore, in addition to our annual goals, Fable 5.1 helped us set quarterly milestones for each of those goals, and it provided 12-month, 90-day , and 30-day action plans. I feel that compared to the output from GPT-6, Fable 5.1 is not only more logical and structured, but its content also aligns better with my expectations and is of higher quality. Also, regarding task time and usage, although GPT-6’s output wasn't very long, it took quite a while. It spent 12 minutes and 29 seconds on this task, costing $ 18.20, while Fable 5.1 finished the same task in just 8 minutes and 40 seconds. Even though it cost $ 10.40, the results were truly excellent. So, for this task , without a doubt, my vote goes to Fable 5.1. For the next scenario, let's see how we can have AI generate sales copy or scripts that match our own tone . This is particularly helpful when we are pitching to clients or need to handle external sales. The example I’m showing here is having two AIs analyze my channel, and then, based on my channel positioning, topics, content , style, and audience needs, generate a high-conversion sales landing page copy for me. Of course, the information I provided wasn’t entirely complete, but I wanted to see what level of performance these two models could reach without having full context. First, GPT-6 did its usual thing, simply placing the written copy directly into the chat box. You can take a look at the content it wrote. Personally, I feel the "AI flavor" is still quite strong, and the logic between many of the sentences doesn't quite flow, while the salesy tone feels a bit too obvious. Now let's take a look at the output from 5.1. First, besides generating the copy for us, it explains in the center of the screen the reasoning behind why it wrote it that way and the logic behind the copy design. As you can see, at the beginning of each line of copy, it tells us the purpose of that sentence —whether it's to open by addressing a pain point, follow up with our core selling points, highlight potential user benefits, or finally, how to alleviate their concerns and drive conversion. After reading the launch letter Fable wrote for me, I felt the content aligned well with standard sales logic, and the phrasing didn't sound overly robotic. This allows me to refine and polish it further based on what it provided. However, for the copy GPT-6 gave me, I might need to ask it to generate a brand new version instead . In terms of task duration and cost, GPT-6 and Fable 5.1 took roughly the same amount of time, with GPT-6 taking 3 minutes and 50 seconds and Fable 5.1 taking 3 minutes and 40 seconds. But in terms of usage costs, GPT-6 still far exceeded Fable 5.1; GPT-6 cost $ 5.60 for this task, while Fable 5.1 only cost $ 3.00. So, considering both the quality of the finished output and the cost, I am still casting my vote for Fable 5.1. For the fourth case study, let's see how well both AIs handle creating presentation slides. First, we uploaded a manuscript from a previous video and asked it to create a six-to-eight-page slide deck for an upcoming speaking event. I also listed the key content for the slides. Since our event is offline, I want the slides to be at the level of a professional business report. Below are some other requirements I set. Let's first look at the results from GPT-6. To begin with, one great thing about GPT-6 is that the color palette it chose for the slides perfectly matches the professional business report style we requested. Additionally, the amount of text on each slide is not excessive, which allows the audience to focus on us during the presentation instead of constantly reading the content on the slides. If you need to make changes to what it created, you can simply select the area you want to modify, click " Request Revision," and enter your instructions. Plus, there are speaker notes included under each slide. Once the slides are ready, besides downloading them, you can also open them in various file formats. The slides produced by Fable 5.1 are quite different from those of GPT-6. First of all, Fable uses a white and orange color scheme. Personally, I feel this style is fine for casual presentations, but for offline business events, these colors might seem a bit too jumpy and lack persuasiveness. Additionally, the slides 5.1 created contain a lot of text, making it easy for the audience to focus entirely on the slides instead of the speaker. Also, if we need to make any changes to the slides, we cannot edit them directly on the slides ; we have to go back to the chat box to submit our revision requests. Regarding task time and token consumption, GPT-6 took 5 minutes and 7 seconds for this task, costing $ 3.80 in tokens, while Fable 5.1 took 4 minutes and 30 seconds , costing $ 6.30 in tokens. Overall, although Fable 5.1 provides much more information in its slides than GPT-6, if I were giving an offline presentation, I would prefer the slides from GPT-6. Therefore, for the task of slide creation, I think GPT-6 has the upper hand this time. We have demonstrated the slide-making effects of two AIs, but I am increasingly finding that in practical work, slides are not as effective as HTML web pages in terms of production cycle or presentation quality. So, if possible, I suggest you try using an HTML web page for your next presentation. So, this time we will have both AIs turn the content of one of my videos into an HTML web page. Here, I am using my previous video on the 14 core concepts of CloudCowork. We are asking both AIs to organize this content into a complete, interactive learning webpage, making it easier for us to quickly and intuitively understand these 14 core concepts. First, let's take a look at the results generated by GPT-6. Overall , the color scheme of GPT6-is quite clear and distinct, and the contrast makes it easy for us to see the changes in hierarchy. When we click to explore the 14 concepts, it takes us directly to the corresponding section. If we click to watch the original video, it can also lead us straight to the YouTube video. When we click on any of these 14 concepts, GPT6-provides us with a detailed explanation of that specific concept. It also tells us which other concepts are related to this one, and informs us of the timeline corresponding to when this concept is mentioned in the video. Here , it generates a table for us, showing how these 14 concepts discussed today work together. To help us understand, GPT6-uses a real-life scenario to connect all these concepts together. I think the production quality of this webpage is excellent, so let's take a look at the webpage made by FB 5.1. First of all, I don't know what font FB 5.1 used here, but the font it presents looks a bit strange. Plus, the combination of green and black against a white background makes the entire page feel less hierarchical compared to GPT6-. Furthermore, when we click to start learning, it simply directs us to the web version of Claude, which is actually quite redundant. Of course, when we click to watch the original video, it successfully leads us to YouTube. The following 14 concept sections are presented in the same way as GPT6-, and when we click on each of these concepts, there are very detailed explanations, along with the timestamps of when the concept was mentioned in the video. Additionally, 5.1 also connects these 14 concepts together to help us understand them. I think what Fable 5.1 does well is that, first of all, if we have never used Cowork before, it provides us with hands-on steps for getting started. Also, at the bottom of the page, there are self-test questions to help us understand all the concepts included. So, overall, I feel that except for its color scheme and font being relatively inferior to GPT6- , the other content and details are actually quite well done. However, I later discovered that the time and token consumption for these two models to complete this task were vastly different. GPT-6 took 6 minutes and 41 seconds, consuming $ 4 in tokens, while Fable 5.1 took a full 17 minutes and 10 seconds and consumed $ 16.20 in tokens. If we consider both task time and token consumption, there is no doubt that I would cast my vote for GPT-6, as it not only delivers good quality but is also much faster and uses fewer tokens. We have now conducted five tests for you. Honestly, after the first three, I really thought Fable 5.1 was a guaranteed winner. But now, after testing the fourth and fifth cases, GPT -6 has caught up, so it is truly hard to call a winner right now. Next, let's test another example. In this upcoming case, we will test the "computer use" capability of both models as they operate the browser. Actually, the examples we just showed you also involved calling the "computer use" function—for instance, when we had the models retrieve data from Google Drive and Gmail. So here, we will have both models use the "computer use" feature to open the design software Canva and create a cartoon portrait suitable for social media based on the profile picture I provided. As you can see here, GPT-6 designed various versions for us, and after three rounds of optimization, it exported this final result. Fable 5.1 also successfully opened Canva, performed two rounds of generation, and eventually produced a cartoon image of Director Li. I think both models did a great job on the avatars, and they both met my requirements quite well. Looking at the production time and token usage, GPT-6 took 8 minutes and 55 seconds for this task, consuming $ 18, while Fable 5.1 took 14 minutes and 30 seconds, consuming $ 12.20. So, you can choose based on your own preferences; if you prioritize speed, choose the GPT-6 model, but if you prioritize cost-effectiveness, you can use Fable 5.1. Finally, let's take a look at the total data. For the six cases we demonstrated today, the GPT-6 model took a total of 39 minutes and 58 seconds, consuming $ 55.20, while the Fable 5.1 model took a total of 51 minutes, consuming $ 52.50. As you can see, their costs are nearly the same, but GPT-6's speed is 20%faster than Fable 5.1. After watching the six cases we demonstrated today, did you notice anything? In my personal experience, I find that for text-only output, Fable 5.1 is clearly better than GPT-6; it not only produces higher quality results and works faster, but it also consumes fewer tokens. This might be related to the fact that they have significantly led other models in writing and text output capabilities. However, once we get into tasks like creating slides, HTML webpages, or calling third-party tools for image generation, GPT-6 really shows its strengths. To summarize, if your tasks are mostly text-based, I personally prefer the Fable 5.1 model. But if your work involves creating presentations or webpages, I recommend GPT-6. That said, I still want to remind everyone that even though GPT-6 is the latest model, GPT-5.6 is still very powerful and sufficient for most knowledge work; using GPT-6 for everything might even be overkill. So, there is no need to use GPT-6 for every task, just as there is no need to use Fable 5.1 for everything. I think they both have their own advantages and areas of expertise, which is why I subscribe to both. The AI market changes so fast that a model that is strong today might be surpassed in a couple of days, so subscribing to both allows you to adjust based on the situation. OK, that concludes my honest evaluation of these two models across six work scenarios; I hope it’s helpful to you. If it was truly helpful, please give this video a like—it helps me a lot. Also, I recently launched a weekly AI newsletter to help you filter out the content in the AI market that is actually worth following; the link to subscribe is in the description and the comment section. Also, if you want to learn more about the GPT-6 model, I happen to have a video that can help you get up to speed and deepen your understanding. If you're interested, you can click here to watch it. See you in the next video, bye!