Opus 5.5 effort levels mostly change graphics polish and edge-case logic, not whether a playable game gets built. One creator gave the same Age of Empires II clone prompt to Claude Code at low, medium, high, xhigh, and max effort, paying from $3.79 to $70 per run, and found the biggest visible gains in rendering and detail rather than core gameplay.
What Opus 5.5 effort levels changed in the same game build
Opus 5.5 effort levels mainly changed graphics quality, settings depth, and edge-case logic in a controlled test where every run received an identical prompt. The test came from Bart Slodyczka's September 2026 video, which asked Claude Code, Anthropic agentic coding tool that runs in the terminal, to build a clone of Age of Empires II at low, medium, high, xhigh, and max reasoning effort.
Each run started from an empty folder on the author's desktop with the same four-sentence prompt: build an Age of Empires II clone with as many resources and as much gameplay as possible. Because the setup was identical, the cost, token count, and build duration of each run are directly comparable, though every figure below is the video author's own measurement rather than a vendor benchmark.
The headline finding: low effort already produced most of the game logic and building variety. What scaled with effort was mostly visual polish, plus fixes for specific bugs like villagers forgetting tasks or getting stuck on terrain.
Cost, tokens, and build time at every effort level
The cost and duration of each build rose steeply with effort, following token consumption. All figures below are as reported by the video author for his own Claude Code sessions.
| Effort | Cost | Tokens | Build time |
|---|---|---|---|
| Low | $3.79 | ~3 million | ~19 min |
| Medium | $9.25 | ~11 million | ~36 min |
| High | $16 | ~30 million | ~60 min |
| Xhigh | $35 | ~80 million | ~1h 41min |
| Max | $70 | ~173 million | ~3 hours |
Two patterns stand out in this table. First, tokens, cost, and time move together, so budgeting for a max-effort run means accepting roughly 18 times the cost and 9 times the duration of the low-effort run. Second, the low-effort build was functional despite being the cheapest, which is why the author concluded that the extra spend mostly bought refinement.
The author also notes the step sizes: medium cost roughly 2.4 times low, high was about 3 times the tokens of medium at triple the price, xhigh was roughly double high in price and time, and max doubled xhigh again. If you can only afford two runs, low and xhigh bracket the practical range.
What low effort already built well
Low effort delivered a playable game with a wider immediate building set than the author expected. For $3.79, the build included a main menu with difficulty options, villagers, buildings, upgrades, and progression to the feudal age.
Specific wins at low effort included the first scout horse unit the author had seen in this series of tests, water on the map, a full roster of buildings including a stable, blacksmith, archery range, siege workshop, and castle, automatic farmer assignment after farm construction, and units that noticed attacks and fought back. Farms also replanted automatically, and villagers moved straight onto a resource node as soon as they finished building next to it. The author rated the low-effort result as genuinely impressive for the price, comparing its graphics favorably to a DeepSeek build he had run the previous week.
The visible problems at low effort were logic bugs rather than missing systems. Villagers periodically forgot their current task, units got stuck near trees and buildings, and workers could not repair damaged structures, a mechanic the author considers important for surviving raids.
What each step up in effort added
Each effort step introduced at least one distinct feature or fix, which is why the author describes the progression as incremental. The gains accumulated roughly as follows:
- Medium ($9.25) doubled the perceived graphics quality, added sheep and wild animals, added game speed and map seed settings, and fixed the stuck-and-forgetting villager bugs. Workers could also repair damaged buildings.
2. High ($16) added civilizations with distinct bonuses (Franks, Teutons, Mongols), convertible sheep you could claim and move, a palisade wall, a wonder victory condition, wild boars, and a military buildings submenu including a monastery. Enemy attacks also arrived in larger, better-timed waves.
3. Xhigh ($35) produced a much richer setup menu with opponent count, fog of war, resource settings, and terrain options, background music, wood-splinter effects while chopping, animated horse hooves, and a full game-to-defeat screen with post-game scores, a first for the author's test series.
4. Max ($70) refined graphics further, added a gate for stone walls, monks, garrison behavior with townsfolk sheltering and firing arrows, and broader research options such as university tech. The author called the result the most complete build but noted the marginal gains over xhigh were smaller than the price jump.
If you only care about one dimension, the author's own summary is that graphics absorbed most of the extra tokens and time. Logic improved too, but the low-effort build already covered the core loop.
Can a master prompt replace high effort?
A detailed master prompt cannot fully substitute for higher reasoning effort, based on the bonus run at the end of the test. The author collected roughly ten requirements he had liked across the five builds, including better graphics, wildlife, functional water with fishing, unit voice responses, a map editor, diplomacy, cheats, monasteries and relics, elevation, and self-testing, then sent all of it to a low-effort session.
That bonus build cost $13.57, consumed about 21 million tokens, and took 1 hour 23 minutes, which places it close to high-effort runs in cost and duration. The result did deliver several requested features: fish in the water, a working map editor with flatten and raise tools, unit voice responses, diplomacy with tribute and alliance offers, a dock with fishing boats, and Age of Empires-style cheat codes like Robin Hood and Cheesesteak Jimmys.
What prompting did not buy was the graphics refinement and gameplay smoothness of high and max builds. The author's conclusion: you can lowball effort and get breadth by specifying requirements, but polish and the details you forgot to specify still require paying for the higher effort tier. That is the practical case for max effort: completeness across everything you did not think to write down.
How to read these results before applying them
These results describe one person's test on one prompt and should be read as such. Every number in this article is the video author's self-reported measurement from his own Claude Code sessions in September 2026, not an independent benchmark and not an Anthropic-published figure.
A game clone is also a graphics-heavy, long-horizon build; cost-to-effort ratios can differ substantially for short scripts, bug fixes, or refactors. If you want to replicate the method, the setup is simple and worth reusing: same empty folder, same prompt, one variable changed at a time, and cost plus token count recorded before evaluating the output.
FAQ
How much did the low-effort Opus 5.5 build cost?
The low-effort build cost $3.79, used about 3 million tokens, and took roughly 19 minutes, according to the video author's own session data. It produced a playable Age of Empires II clone with most buildings and core gameplay logic.
Was max effort worth 18 times the price?
The max build cost about $70 versus $3.79 at low effort. It delivered the best graphics, gates, monks, and garrison behavior, but the core game was already playable at low effort, so the premium bought polish and completeness rather than basic capability.
Did a detailed prompt match higher effort levels?
Only partially. The master-prompt run at low effort cost $13.57 and delivered many requested features like fishing, a map editor, and diplomacy, but it did not reach the graphics refinement or gameplay smoothness of the high and max builds.
What was the biggest difference between effort levels?
The video author identified graphics as the single biggest difference, with gameplay logic fixes such as stuck or forgetful villagers as the second. Building variety was broad even at low effort.
How long did each effort level take to build?
About 19 min at low, 36 min at medium, 60 min at high, 1h 41min at xhigh, and roughly 3 hours at max. Build time rose in near-lockstep with token consumption, which drove the cost curve.
What did xhigh effort add over high effort?
Xhigh added a much deeper pre-game setup (opponents, fog of war, resources, terrain), background music, richer effects such as wood splinters while chopping, and a proper defeat screen with post-game scores. The trade-off was $35 and about 80 million tokens versus $16 and 30 million.
Could villagers repair buildings at low effort?
No. Repair was missing entirely at low effort and appeared at medium. The author treats repair as essential for surviving raids, so this single fix is one of the clearer reasons to leave the lowest tier.
Are these costs official Anthropic pricing benchmarks?
No. All cost, token, and duration figures are the video creator's self-reported measurements from his own Claude Code sessions, not vendor-published benchmarks.
Can Claude Code run this same comparison on other prompts?
Yes, and the method travels well: fix the folder, fix the prompt, change only the effort level, and record cost, tokens, and duration before judging the output. The harder part is scoring quality objectively, which is why the author narrated specific checks like repair, sheep conversion, and victory conditions.
Turn your own video tests into a written comparison
A structured comparison like this one lives or dies on the details: exact costs, token counts, and what changed at each step. That format has been refined by the Brazilian dev community, from Gustavo Dev Doido's Bootcamp do Dev Doido to the Crazystack Typescript community around crazystack.com.br, and the same structure works for any model comparison, tool review, or build walkthrough you have already recorded.
Skala Blog does exactly that: paste a YouTube URL, get a transcription, and generate a well-structured article from what you already said on camera. Your five-effort-level test, or whatever your version of it is, deserves a written home.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits