Blog Article
Fable 5.2 vs Opus 5.5: I Tested Every Effort Level on the Same Prompt
I ran Fable 5.2 and Opus 5.5 on every effort level with the same paintball game prompt, then played each result. Here's which model I'd pick at each level and what
I wanted to know one thing: when should you pick Fable 5.2 over Opus 5.5, and does the effort level change the answer?
So I gave both models the same prompt on every effort level (low, medium, high, extra high, and max) and played every result.
The short version: on low and medium, I'd still choose Opus. Once we cross into high, Fable takes over. Fable on max was the clear winner, and it was also the most expensive run on the board.
How the test stayed fair
I ran this in ClearBench, the in-house benchmarking tool we built. Every lane gets the same setup:
- The same prompt. One prompt, pasted once.
- A clean start. No memory, no saved sessions, no skills, plugins, MCP servers, or hooks. No CLAUDE.md, no AGENTS.md, no markdown files at all.
- The same toolbox. Read and write files, shell with internet, web search, and web fetch.
- The same start time. Every effort level kicks off at the same moment, each in its own sealed folder, so there's no cross-contamination between tests.
Tokens are tracked from the CLI's own report and priced off the public API listings.
The prompt
The gist: build an ultra realistic first-person paintball game for the browser called Off the Break.
The full prompt has a lot more detail on how the game should operate. If you want to copy it, you can screenshot each paragraph on screen at [01:31].
One thing I didn't get specific about was how the characters should look. That shows up in the results, so keep it in mind.
What went wrong along the way
I'm not going to hide the messy parts.
- The first set of runs failed and burned through tokens. By the end I had been recording for five hours.
- The runs took far longer than I expected. Three hours in, max on both models still had no preview ready, so I started testing the others.
- Opus on max never made it. While I was waiting on it, the timer reset. I think the run failed and I accidentally hit restart. Either way, I couldn't play it and I have no cost for it.
For this test I wasn't focused on speed anyway. I was focused on quality of output and cost.
Opus 5.5, level by level
Low. The sound effects were questionable, but there was a nice paint splash on the goggles and I loved the call outs. I wasn't that impressed, but what would we expect from low effort, right?
Medium. A very basic looking environment. Sliding worked. The hit boxes didn't seem to be working, because I know I was hitting that player. Not horrible. Not the best.
High. I got a clear hit this time, and it added a "G to wipe your lens" control. The characters looked okay. I'm sure a detailed PRD would have done a lot better here. Still, it's a playable video game.
Extra high. A good slide with real movement to it, and I liked the realism on the bunkers. This was my favorite Opus run.
Max. Bugged out. Nothing to test.
Fable 5.2, level by level
Low. It didn't give me an inverted mouse option, which threw me off. It wasn't that big of an improvement over Opus. If I'm being brutally honest, I would choose Opus on low over Fable on low. Give me Opus nine times out of 10.
Medium. The running movement felt very realistic. I didn't like the sound effects, and the bunkers looked lazily made. Not impressed again. I would have chosen Opus.
High. Now we're getting somewhere. It looked a lot better right away. When I leaned left, the weapon switched to the left side of the body. That's nice attention to detail, and significantly better than low and medium. You can see it at [06:10].
Extra high. I could see my stats, the bunkers looked realistic, and the slide was nice. At that point it was my favorite, hands down.
Max. The player looked so much better, and it had a wipe lens feature. I was impressed by this one. With some fine tuning it would look even better. Fable on max is the clear winner.
What each run cost
Here's the scoreboard as I read it off:
- Opus on low was the cheapest run
- High was the second cheapest, then medium
- Then Fable medium, then Fable low
- Opus extra high: $25
- Fable 5.2 high: $29
- Then up to $38 on extra high
- Almost $45 on max
- Opus max: unknown, since it bugged out
One interesting discovery: medium and low cost roughly the same amount, and those were the outputs I didn't like.
The pattern is hard to miss. The runs I liked best sat at the top of the cost list.
Lesson learned: don't run 10 sessions on one small server
The time was the shocker for me.
The server I ran this on has around 16 cores and 32 gigs of memory allocated to it. With five effort levels on two models, that's 10 parallel sessions pulling from the same machine at once. I have a feeling that's why everything ran so much longer than it would have one by one.
Moving forward, I'm running these benchmarks on my daily driver instead.
So which one should you pick?
Based on this one test:
- Low or medium effort: Opus 5.5. The Fable results at those levels didn't beat it.
- High and up: Fable 5.2. That's where the attention to detail started showing up.
- Best result, period: Fable 5.2 on max, as long as you're okay with the cost.
Two caveats. This was one prompt, and it wasn't specific about things like character design. A detailed PRD would likely have lifted every run. And I never got to compare max against max, because Opus didn't finish.
I'm not an AI expert. I'm building in public and sharing what works. There's a lot more Fable 5.2 testing coming, so if you want a specific build tested, drop a comment on the video with who you are, what you do, who your audience is, and your question. I'll add it to my queue.
Watch the full test and see every game in action: https://www.youtube.com/watch?v=3WsHeceptoQ
Watch the full walkthrough on YouTube.
Watch on YouTube