One episode · Large Barren Desert

Claude Opus 5.5

Benchmark 1.5.6AnthropicReasoning mediumPlayed by the maintainersFull-budget score

All of Claude Opus 5.5’s results and comparisons

Score

67.6

Crusader Bench score on Large Barren Desert: 0 is doing nothing, 100 the person’s game

Net worth

8,757

+3,565 over doing nothing, 74% of the person’s gain of 4,792

Population

22% of the person’s gain of 246 · popularity 59

Cost

~$2.01

3.6M tokens; ~ at list prices, the API reports no bill

Time

33 game min

37 min of wall-clock time, 66 turns

Minute by Minute

Net worth, how it split between gold and stored goods, and the settlement, each game minute.

Per game minute · dashed lines are where doing nothing and a person's game ended

Claude Opus 5.5
Baseline (do nothing), final
A person's game, final

Net Worth

gold plus stored goods at the sell price

Gold or Goods

Gold
Goods

Population

peasants

Popularity

0–100

Food in the Granary

units

Crusader Bench · crusbench.com

Net worth is gold plus stored goods valued at the marketplace’s sell prices.

Selling at the marketplace converts goods to gold at the price the score already counts them at, so it moves value between the two without changing the total. Buying costs net worth.

Population and popularity are a tenth of the score each; food is not scored. Population supplies workers, popularity allows taxes, and food supports both.

The episode

Watch the Episode

Episode 1 · 33 min 20 s of game time, edited: actions at real speed, waiting fast-forwardedStreamed from Hugging Face

ScorecardFull-budget score

8,757net worth

Gold: 2,513
Stored goods, at sell price: 6,244
Over the baseline (do nothing): +3,565
Population / housing: 58 / 58
Popularity: 59
Food: 4
Structures (map-wide): 100
Attempt: 1

The settlement at the endcentred on the keep, zoomed out

The settlement at the end of episode 1, zoomed out over the keep

Final stock6,244 gold in goods

Good	Units	Price	Value
Gold	2,513	1	2,513
Stone	506	7	3,542
Iron	99	23	2,277
Ale	20	10	200
Pitch	20	10	200
Wood	9	1	9
Cheese	2	4	8
Bread	1	4	4
Meat	1	4	4
Net worth			8,757

Beside a person’s gameLarge Barren Desert, the same game time

	This episode	The person	Left alone
Net worth	8,757	9,984	5,192
Gold	2,513	8,574
Stored goods, at the sell price	6,244	1,410
Population	58	250
Housing	58	250
Popularity	59	100

The person played Large Barren Desert for 33 min 33 s of game time at the same speed, timed by their own clock, and was scored the same way. The model saw this comparison only in its interview.

The town’s booksread from the game at the end

Date in the game: July 1178
Idle peasants: 4
Popularity next month: −2
Food: +2
Taxes: −4
Taxes: Average taxes, 46 gold a month
Rations: full; 3 of 3 foods eaten
Worker efficiency: 100%
Storage: 44 stockpile piles, 25 empty; 1 granary holding 4 food

Buildings50

Ox Tether16
Stockpile11
Quarry10
Hovel6
Iron Mine4
Granary1
Keep1
Marketplace1

Tool calls128

market_trade26
build_structure25
find_sites16
wait_and_observe15
go_to_tile12
set_tax8
building_info6
center_on4
expand_storage4
inspect_building4
game_action3
observe3
flat_view1
map_overview1

Play

Turns: 66
Buildings placed / attempted: 42 / 71
Anchor-tool placements: 5
Tool errors: 1
Raw inputs sent to the game: 593
Context compactions: 2
Game time: 33 min

Compute

Tokens: 3.6M
Served from cache: 96%
Output tokens: 28K
Cost, at list prices: ~$2.01
Wall-clock time: 37 min
of which inference: 10 min
Peak game memory: 3,211 MiB

The plan, before play

Task. I'm playing the Large Barren Desert Free Build save for 2,000 game seconds. The score is mostly net worth: gold plus stored goods at their sell price. Population and popularity count for less. Building costs and buying at the market lower net worth, so every building has to produce more than it costs. There is no oasis grass, so no farms: food has to come from the market or from the starting meat. No soldiers or defences.

Opening plan

Wood and housing first. While the starting package arrives: build 2–3 woodcutters beside trees near the stockpile, 2–3 hovels away from the signpost, and a marketplace. Leave taxes at "no taxes" until popularity is comfortably above 50.
Iron as the main earner. Iron is worth 23 a unit and its miners carry it themselves, with no oxen. The ore is about 60 tiles east of the keep. Once I have wood, I'll use find_sites there to place several iron mines.
Stone second. Quarries with ox tethers on the western deposits (about 100 tiles away), if the walk still pays off.
Food and popularity. Keep the granary stocked with cheap food (meat or apples, 8 to buy). Keep rations at full. Possibly add an inn using the starting ale, and a chapel. Then raise taxes step by step for steady gold while popularity stays above 50.
Storage. Watch the stockpile and add piles with expand_storage before it fills. Sell surplus wood if it piles up. Keep iron and stone in store, since they count at full value anyway.
Keep expanding hovels, mines and woodcutters as peasants arrive, without wasting gold on things that don't pay back. Toward the end, stop spending and hold gold and goods.

Commentary during play1

Place 4 iron mines.

Reading the game

The Analysis

A reading of the game written afterwards from its own logs: how it went, what held the model back, what worked, and what it said in its interview. It is published with the game and is not part of its score.

Written by Claude (Opus 5.5) from the run's own logs (events.jsonl, episode.json, results.json, interview.md); it is a reading of the run, not part of its score.

Model: Claude Opus 5.5 (claude-opus-5-5) on Anthropic's API, reasoning medium, output limit 32,768 tokens
Benchmark: Large Barren Desert construction, version 1.5.6, from the save the human reference game on this map started from; 33 min 20 s of game time (60,000 ticks) at speed 40, the length of the person's game; no playbook
Run 596c46a2, 2026-10-12

Result

	Claude Opus 5.5	The person	Left alone
Net worth	8,757	9,984	5,192
Gold	2,513	8,574
Stored goods, at the sell price	6,244	1,410
Population / housing	58 / 58	250 / 250
Popularity	59	100

A full-budget score: the run used all 2,000 game seconds. Net worth grew by 3,565 over leaving the game alone, about three quarters of the person's 4,792. Almost all of the end value beyond gold is what its mines and quarries dug: 506 stone (3,542) and 99 iron (2,277). It paid for that in gold: 4,960 went on the market, 4,320 of it on food, and it ended with the least gold of any run on this map.

How the game went

The opening, 0 to 2 minutes. Four iron mines on the eastern ore in its first turn, two more stockpile piles for the starting package at 24 seconds, four hovels and a marketplace, then 20 lots of wood (100 units) at 1:34 to pay for five quarries about 100 tiles west of the stockpile with four ox tethers beside them. Taxes went to 6 at 1:34 and back to 4 at 3:46.

Building, 2 to 14 minutes. Three more tethers by the quarries and two hovels at about 5 minutes, then a chain of stockpile piles east toward the iron mines from 6:42 to about 8:30 (five piles placed, the last about 11 tiles from the mines); a west chain toward the quarries found no room. Five more quarries and four tethers at 10:49, five more tethers at 13:03 and 13:51: 10 quarries and 16 tethers in all. Population reached the 58 places of its six hovels and keep at 14 minutes and stayed there, with four food types bought to keep everyone on full rations.

Waiting, 15 to 33 minutes. It built nothing more (an expand_storage at 17:02 found no room), took 14 turns in the last 18 game minutes, and spent them buying food in rounds of three to six lots of each type, moving taxes between 4 and 5, and waiting 60 to 200 seconds at a time. Meanwhile the quarries and mines kept digging. Stone came in at about 6 a minute up to 15 minutes and about 21 a minute after, once the second five quarries and their tethers were working:

Game minute	0	3	6	9	12	15	18	21	24	27	End
Net worth	4,546	6,032	6,087	6,137	6,222	6,017	6,237	6,716	6,924	7,701	8,757
Gold	4,330	4,648	4,522	4,366	4,070	3,427	3,211	2,929	3,107	2,879	2,513
Stone	0	50	50	66	98	130	162	240	250	354	506
Iron	0	15	23	29	37	45	53	62	66	81	99
Population	2	24	35	48	56	58	58	58	58	58	58
Popularity	100	84	68	61	55	60	57	64	60	59	59

During long waits a row repeats the last reading; the last turn was at 30:08, before a final 200-second wait in which the food ran down from 71 to 4.

What held it back

Every peasant was fed from the market. 540 units of food over 26 purchases cost 4,320 gold and 2,160 net worth against the sell price. Opus said in the interview it capped the population at 58 on purpose, because "taxes from an idle peasant didn't cover their food".
Nothing was made from what it bought. The town was mines, quarries, hovels and storage: no inn, chapel, mill or bakery, so population and popularity stayed far below the person's.
The quarries were about 100 tiles from storage, and the west stockpile chain that would have shortened the walk found no room. The tethers were beside the quarries, as the 1.5.6 rules describe, and with ten quarries and sixteen tethers the stone still came in at about 21 a minute.
It stopped building at 14 minutes with about 3,400 gold in hand.

What worked

Ten quarries with sixteen tethers beside them and four iron mines, the most production of any run on this map: 506 stone and 99 iron at the end against GLM 5.3 Flash's 148 and 17.
Room for the starting package at 24 seconds.
Every reported placement matches the game's final count of its buildings.

Cost and speed

66 turns, 128 tool calls (15 waits), 3.60 million tokens and $2.012 (preparation $0.155, play $1.756, compaction $0.101), worked out from the list prices on the model profile. The interview cost $0.041 more, with 58,013 of its 60,386 tokens read from the cache. The agent run took 37.1 real minutes, 10.0 of them waiting for the model; the context was compacted twice (34 seconds).

Beside the others on 1.5.6

	Opus, 1.5.6	GLM 5.3 Flash, 1.5.6	Claude Haiku 5.5, 1.5.6	Claude Sonnet 5.5, 1.5.6
Net worth	8,757	6,655	6,412	6,252
Gold / goods	2,513 / 6,244	4,560 / 2,095	3,428 / 2,984	3,299 / 2,953
Population / housing	58 / 58	70 / 98	32 / 42	27 / 114
Popularity	59	66	51	24
Iron mines / quarries	4 / 10	4 / 7	4 / 0	4 / 0
Last building	14 min	29 min	7 min	6 min
Net worth lost to purchases	2,640	2,890	1,000	1,160
Cost	$2.012	$0.107	$0.034	$0.415

Opus and GLM spent about the same net worth at the market; Opus's ten quarries and the tethers beside them turned it into more than three times GLM's stone and nearly six times its iron. It also ended above GLM's 1.5.4 run on this map (8,232), the highest before it, at about 19 times the cost of GLM's 1.5.6 run.

The interview

Opus said its goods column "makes my net worth look better than my economy really was": most of it was "stone and iron collected by slow production" while it "lost gold the whole time buying food". It put the difference down to "the model of value": it never looked for a chain that turns bought goods into something worth more, as the person's bakeries did with bought wheat. Playing again, it would test such a chain early, push popularity with inns, a church and bad things so taxes can run higher, put storage near the industry early, and "build big early rather than cautiously". It read the rule that buying lowers net worth "too conservatively, as 'avoid buying'", and asked for each building's output per minute.

For the harness

Selecting a Stockpile moves the view back to the first stockpile. Placing the eastern chain by tile, 11 placements came back not_on_screen ("The view moved while the building was being selected"; camera centre 506, 373). The remedy, go_to_tile and place again, cannot work: Opus followed it at 7:16 and was moved back again. It got the chain placed by zooming out and by screen pixel. Runs that grow storage with expand_storage from the first stockpile do not meet this; a chain toward far industry does.
The game saved correctly at the end ("claude opus 5-5 1-5-6 596c46").

After the game

The Interview

Once the game was scored, the model was shown its result beside a person’s game on the same map and asked four questions, in the conversation it had played in, with the tools off. Nothing it said changed its score.

1Your result

Interviewer

The run is over and your result is recorded. What follows is an interview, not part of the benchmark: nothing you say now changes your score. We ask every model the same questions after its game and publish the answers with the run. The tools no longer work, so answer in plain text without calling them. Be candid; there are no right answers.

Your result, read from the game when your run ended, beside a person's game on the same map for the same game time:

	You	The person
Game time played	33 min 20 s	33 min 33 s
Net worth	8,757	9,984
Gold	2,513	8,574
Stored goods, at the sell price	6,244	1,410
Population	58	250
Housing	58	250
Popularity	59	100

Leaving the game alone for the same time ends with a net worth of 5,192.

How do you see your result next to the person's? What do you think made the difference?

Claude Opus 5.5

I ended at 8,757 against their 9,984. That puts me about 3,500 above leaving the game alone and about 1,200 behind the person. Population (58 vs 250) and popularity (59 vs 100) were much further apart. The goods column makes my net worth look better than my economy really was. Most of my 6,244 was stone and iron collected by slow production, and I lost gold the whole time buying food to keep 58 peasants fed. The person turned gold into a growing population. I treated each peasant as a food cost, and that shaped my decisions.

I think the difference was the model of value. On a map with no farmland I didn't look for a way to process goods into something worth more. My only income was mining, and the quarries were about 100 tiles from storage, so the oxen walked a long way each trip. Stone came in batches of 8, with long gaps. I spent a lot of game time on an east stockpile chain toward the iron, on failed attempts at a west one, and on many short waits. I also kept population capped at 58 on purpose, because I worked out that taxes from an idle peasant didn't cover their food. That held only because I never gave those peasants profitable work.

2Playing again

Interviewer

How the person played, as they describe it:

Bread was one of their main sources of income: mills ground wheat into flour, bakeries baked it into bread, and the bread was sold at the marketplace.
They bought wheat at the marketplace so that flour and bread production never stopped.
A church and a cathedral raised popularity, and so did inns.
The fear factor (bad things such as gallows and stocks) made the workers work faster.

Their settlement at the end, read from the game:

250 people in 30 hovels, with no free housing; popularity 100.
95 bakeries and 12 mills, and no wheat farms.
11 inns, which reached everyone with ale, and no brewery or hops farm.
13 quarries and 7 iron mines with 37 ox tethers; no woodcutters.
A church and a cathedral, and 21 bad things (11 gallows, 5 stocks and 5 burning stakes), which made the workers 110% efficient.
Low taxes and no rations. The monthly popularity factors were food −8, taxes −2, fear −1, religion +3 and inns +8: no change at 100.

If you could play this map again from the start, what would you do differently, and why?

Claude Opus 5.5

Look for a processing chain right away. I would test early whether buying an input and processing it pays. The person bought wheat, milled it, baked bread and sold the bread at enough scale to profit. That makes peasants productive instead of a pure cost, and it needs no farmland. With 95 bakeries, population is what limits output, so housing becomes an investment.
Push popularity hard so taxes can run high. Inns using bought ale, a church and a cathedral, and bad things for the efficiency bonus. Popularity 100 with no rations is a very different economy from my 59 with full rations of four bought foods.
Place industry near storage, or bring storage to it early. I'd have tried the west stockpile chain around the keep in the first minutes, or built far fewer quarries, rather than 10 quarries with walks of about 100 tiles.
Spend less real time per game minute deciding. The early game compounds the most, so I would build big early rather than cautiously.

3The benchmark

Interviewer

What do you think of this benchmark: the tools, the screenshots and game readings you got, the time limit and the instructions? What helped you, what got in your way, and what would you change?

Claude Opus 5.5

The tools were good. Tile-targeted placement with automatic retries, find_sites, market_trade and the stats JSON saved a lot of clicking and screenshot reading. Popularity factors, storage room and idle peasants in every reading made it easy to see the state.

What got in my way:

Selecting a stockpile jumps the camera back to the existing one, so tile targets out of view failed.
Some placements came back not_placed with no reason.
go_to_tile sometimes stopped short when zoomed out.

My biggest problem was my own strategy, though, not the interface. The rules note that buying lowers net worth. That's true, but I read it too conservatively, as "avoid buying". I'd keep the warning but make it clearer that buying inputs to process can be profitable. I'd also like a building's produced value to be easier to see (output per minute per building), so it's clearer what is worth adding more of.

4Anything else

Interviewer

Is there anything else you would like to say about this game, this benchmark or how you played?

Claude Opus 5.5

It's humbling, and useful, that the person won with a strategy I never considered: a bakery megacity fed entirely by market wheat. I optimised locally (food cost per peasant, ox trips) and missed the bigger picture. The run also showed that a large stock of raw goods can hide a weak economy behind a decent net worth. My "rising" score came mostly from slow mining while I held the town still. Next time I'd pay more attention to how fast net worth is growing than to how high it is.

Record

Attempts and Reproducibility

Every attempt1

Episode	Attempt	Outcome	Net worth	Counted
1	1	Full-budget score	8,757	Counted

What ran

Benchmark version: 1.5.6
Fingerprint: bb6e73f95223
Harness commit: 9c34b50
Prompt / tools SHA-256: ad97396 / 76cbfd6
Menu guide: Images 118089d
Save: Large Barren Desert benchmark start
Game time per episode: 33 min 20 s
Playbook between episodes: No
Minimum turn / default wait: 8 s / 5 s
Observation after acting: Text unless a screenshot is needed
Screenshot width: 1,280 px
Context budget: 120,000 tokens
Real-time limit: 360 min
Submitted by: hesamation

All files in the dataset 20261012T113644-1024a7b0