The Marketing Ninja · 2026-10-30 · 5 min
Key moments - from our scoring
Substance score
26 / 100
Five dimensions, 20 points each
Speaker B conducted an extensive test with staff member Jules comparing 16 variations of Claude models and effort settings for video production, specifically for content created with Cesar Milan's Better Dog Supplements brand. The core finding: Opus 5.5 Low delivered quality scores of 4.4 at $0.07 per request in 32 minutes, while Opus 5 High achieved 4.6 quality at $0.28 (four times the cost) in 69 minutes. The quality difference - slightly better audio and music in the premium version - is imperceptible to most audiences. This testing reveals that model selection (Opus, Claude variants) and effort settings (Low, High, Ultra) create massive cost multipliers, potentially saving organizations 75% in token spend without noticeable output degradation. The principle applies beyond video to website building, funnel creation, and ad production. For operators managing significant AI spend (the host exceeded $20,000 in September alone), understanding these trade-offs directly impacts margins.
Hosted by Manuel Suarez. The show scores 42.5/100 on substance across its analysed episodes. See the full show profile.
Opus 5.5 Low costs $0.07 per request while Opus 5 High costs $0.28 - approximately 75% less - with imperceptible quality differences to audiences (4.4 vs 4.6 out of 5).
Opus 5.5 Low produced video in 32 minutes while Opus 5 High took 69 minutes, more than doubling production time for minimal quality gains.
No - Speaker B found that quality differences in audio and music between lower-cost and premium settings are not noticeable to the human eye and not worth the 4x cost multiplier.
Test simpler settings on ordinary tasks first, give both versions the same brief, compare usable results, use more power only when the job needs it, and always test small samples before scaling workflows.
Over $20,000 in September alone, which prompted the systematic 16-way testing to optimize cost and identify redundant spending on premium model settings.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode contains a handful of concrete cost-optimization data points (0.07 vs 0.28 cents, 32 vs 69 minutes, quality delta of 4.4 vs 4.6) but the core insight - use a cheaper model tier for routine tasks - is thin and occupies the entire 5 minutes. The closing rules are generic platitudes that add nothing beyond the one central observation.
when we use Opus 5.5 low, the quality was 4.4 and the estimated plan share was 0.07. And it took 32 minutes to produce the video
the Same exact request, Opus 5, not 5.5, but high, went to 28 cents. Slightly higher quality, and it took more than double the time
The sole argument - lower-tier AI settings are cheaper and often good enough - is entirely expected cost-optimization logic that any practitioner would intuit without testing. There is no contrarian framing, no first-principles reasoning, and no counterintuitive conclusion; the episode confirms the obvious.
if you go opposite, 5.5 ultra at all times is going to cost you a lot more than if you go low when you don't need to go high
those little subtle changes are not really visible to the audience. But you're going to save a lot of money in the long run
There is no guest - the episode is a solo monologue by the host. The host demonstrates some real-world AI spending ($20k/month) and an influencer partnership with Cesar Milan, but the episode itself reveals little practitioner depth beyond running a basic two-condition comparison.
I spent over $20,000 on AI use in the month of September. That's the reality.
Cesar Milan is one of my partners. We, uh, built a brand together called Better Dog Supplements.
A handful of real numbers (cost per run, production time, quality score) give the episode some grounding, but the claimed 16-test methodology is almost entirely absent from the transcript and the metric 'estimated plan share' is never defined, leaving the evidence framework murky.
the quality was 4.4 and the estimated plan share was 0.07. And it took 32 minutes to produce the video
we're paying 0.07 cents, which is a quarter of the cost
There is no conversation - the episode is a monologue with embedded audio clips of a dog-care video. No questions are posed, no claims are challenged, and there is no interlocutor to push back or probe the methodology.
who could tell the difference between one or the other?
you be the judge.
Computed from the transcript - who did the talking, and the words that came up most.
I spent over $20,000 on AI in September, so I had my team test the same video production with Claude 16 different ways. One setting produced the video at a quarter of the cost, and you cannot really tell the difference. Jules on my team ran the same request through different Claude models and effort settings. Opus 5.5 low scored 4.4 out of 5 on a blind quality score, at an estimated plan share of 7 cents, in 32 minutes. Opus 5 high scored 4.6, came in at 28 cents and took 69 minutes. You watch both versions and judge for yourself. Then I share my rule for choosing a model: try a simpler setting on an ordinary task, give both versions the same clear brief, compare the usable results including corrections, and use more power only when the job needs it. Want to build AI systems for your business with my team? Join AI Transformation Plus: About Manuel Suarez: Manuel Suarez is the Marketing Ninja, author of Marketing Magic, and CEO of Attention Grabbing Media (AGM). Follow Manuel Suarez on Social Media: Facebook: Instagram: TikTok: X: LinkedIn: Learn More About AGM: Need Help with Your Marketing? Marketing Magic by Manuel Suarez:
Transcribed and scored by The B2B Podcast Index.
Speaker A: Stress on a dog before people notice it, like the silent stress.
Speaker B: So that is the 0.28 version of that video.
Speaker A: Stress on a dog before people notice it, like the silent stress. Uh,
Speaker B: and we're paying 0.07 cents, which is a quarter of the cost. Who could tell the difference between one or the other? I spent over $20,000 on AI use in the month of September. That's the reality. We just did some testing because of my psychotic spending. So I have one of my staff members, her name is Jules. Uh, I had her test the same video, the same video production, 16 different ways. And what we found is absolutely revolutionary. Um, and I want to pass that along to you right now. Cesar Milan is one of my partners. We, uh, built a brand together called Better Dog Supplements. And we have a big organic following and social media following. And so different models and effort settings. Remember, there's two things the model and the effort setting. You can go opus5.5 ultra code, right, which is, uh, the hardest, most intense and most advanced effort model. Or you can go 5.5 low. Believe it or not, it's a massive difference. And if you do, if you go opposite, 5.5 ultra at all times is going to cost you a lot more than if you go low when you don't need to go high. This is important. I'm not talking about micro money, I'm talking about macro. This is like four or five times the money. So blind quality scores out of five. So time and estimated plan use recorded. For example, when we use Opus 5.5 low, the quality was 4.4 and the estimated plan share was 0.07. And it took 32 minutes to produce the video. Uh, the same thing applies to building websites, funnels, campaigns, advertising videos, you name it. And then the Same exact request, Opus 5, not 5.5, but high, went to 28 cents. Slightly higher quality, and it took more than double the time to get it done. 69 minutes versus 32. So 75% lower estimated plan share in this comparison. Five models, one video. This is exclusively a test on the cloud models, but you got to keep in mind that the same rules apply to any models. There's going to be massive differences in cost. The same exact request we're going to go to over Here, it says OPOS 5.5 low. Actually, this is, um, interesting because it's a higher model, right? Instead of Opus 5 High, we're going to Opus 5.5 Low, and we're paying 0.07 cents, which is a quarter of the cost. Now, the system claims that it's 4.4 in quality versus 4.6. And you tell me, you be the judge.
Speaker A: Stress on a dog before people notice it. Like the silent stress, isolation, uh, it's a silence. Silence is healthy. Silence is good. But when they're quiet, when they're quiet and tense,
Speaker B: who could tell the difference between one or the other? Aside from, like, the music being different, who can tell the difference here? It's interesting, right? Yes, it is higher quality, but is very, very not noticeable to the masses. To the point that it is four times more expensive. It's not really worth it. The audio right on that, uh, video was maybe slightly lower in quality. The music was slightly lower. But to the human eye, you cannot really tell the difference between one or the other. So you got to keep that in mind. Like, you got to understand that those little subtle changes are not really visible to the audience. But you're going to save a lot of money in the long run. My rule for choosing a model one, try a simpler setting on an ordinary task. Two, give both versions the same clear brief. Three, compare the usable results, including corrections. Four. Use more power when the job needs it. As you start spending more and more money, this becomes more and more important. You can say something like, can a lower cost setting do this job well? Test a small sample before changing the workflow. That in the long run will be very important for you to track in. When it comes to your AI use.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
544: Repositioning NutanixRenegade Marketers Unite · on Token cost optimization77 / 100
S2. Ep29 - Set Up Claude Once, Use It Everywhere: Projects, Skills & Cowork ExplainedAI Automations for Business · on Claude Opus 576 / 100
Hedaux , Hogg & The Machines 1ReThink Productivity Podcast · on Claude Opus 5
