One episode · Large Barren Desert

GPT-6 Luna

Benchmark 1.5.6OpenAIReasoning mediumPlayed by the maintainersFull-budget score

All of GPT-6 Luna’s results and comparisons

Score

45.9

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

Net worth

7,354

+2,162 over doing nothing, 45% of the person’s gain of 4,792

Population

106

41% of the person’s gain of 246 · popularity 57

Cost

$0.13

6.6M tokens

Time

33 game min

58 min of wall-clock time, 143 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

GPT-6 Luna
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

7,354net worth

Gold: 1,246
Stored goods, at sell price: 6,108
Over the baseline (do nothing): +2,162
Population / housing: 106 / 130
Popularity: 57
Food: 8
Structures (map-wide): 124
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 stock6,108 gold in goods

Good	Units	Price	Value
Gold	1,246	1	1,246
Stone	530	7	3,710
Iron	100	23	2,300
Ale	7	10	70
Meat	7	4	28
Net worth			7,354

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

	This episode	The person	Left alone
Net worth	7,354	9,984	5,192
Gold	1,246	8,574
Stored goods, at the sell price	6,108	1,410
Population	106	250
Housing	130	250
Popularity	57	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: 24
Popularity next month: −11
Food: −4
Taxes: −12
Religion: +3
Ale from the inn: +2
Taxes: Extortionate taxes, 148 gold a month
Rations: half; 1 of 1 foods eaten
Worker efficiency: 100%
Storage: 28 stockpile piles, 12 empty; 2 granaries holding 7 food

Buildings84

Ox Tether39
Hovel15
Quarry12
Stockpile7
Iron Mine4
Granary2
Chapel1
Church1
Inn1
Keep1
Marketplace1

Tool calls172

find_sites37
build_structure25
go_to_tile20
market_trade20
building_info17
set_tax11
status8
wait_and_observe7
place_near6
get_inventory4
inspect_building4
map_overview4
expand_storage3
observe2
update_plan2
center_on1
set_rations1

Play

Turns: 143
Buildings placed / attempted: 68 / 73
Anchor-tool placements: 13
Tool errors: 2
Raw inputs sent to the game: 697
Context compactions: 3
Game time: 33 min

Compute

Tokens: 6.6M
Served from cache: 97%
Output tokens: 73K
Cost, as billed: $0.13
Wall-clock time: 58 min
of which inference: 29 min
Peak game memory: 3,231 MiB

The plan, before play

I’ll grow the settlement’s economy for the full time budget, prioritizing productive buildings, housing, population, and popularity while avoiding all military recruitment and defenses. I’ll first check the live map and resources, then build housing and available production near useful terrain and storage. I’ll monitor staffing, supplies, and popularity as the town grows, expanding storage or adjusting taxes and rations as needed. I’ll use the desert’s stone and iron deposits for production; with no farmable oasis, I won’t spend resources on a food chain that cannot work.

Commentary during play0

The model played this episode without writing commentary between its tool calls.

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: GPT-6 Luna on OpenRouter, providers openai then azure, 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 f2a35309, 2026-10-12

Result

	GPT-6 Luna	The person	Left alone
Net worth	7,354	9,984	5,192
Gold	1,246	8,574
Stored goods, at the sell price	6,108	1,410
Population / housing	106 / 130	250 / 250
Popularity	57 (−11 next month)	100

A full-budget score: the run used all 2,000 game seconds. Net worth grew by 2,162 over leaving the game alone. Nearly all the stored value is what its twelve quarries and four iron mines dug: 530 stone (3,710) and 100 iron (2,300). It spent 6,780 gold at the market, 5,040 of it on 630 units of cheese and meat, and ended on half rations and extortionate taxes.

How the game went

The opening, 0 to 5 minutes. Six hovels against the keep and three more stockpile piles in the first minute, four iron mines on the eastern ore at 41 seconds, a chapel and an inn by the keep at 1:24, then eight quarries on the western stone, about 100 tiles from the stockpile, with an ox tether beside each by 4:18. A marketplace went up at 3:36, beside the quarries, after a wood purchase found none.

Growing, 5 to 15 minutes. Six more hovels, three more stockpile piles and a granary by 8:31, two more quarries at 8:31 and two at 11:50, a church at 13:21 and three more hovels at 14:48. Food came from the market, in lots of 100 units of cheese or meat at first and 20 to 25 later. Population reached 92 at 15 minutes with popularity at 90, but gold fell from 4,340 to 1,557 and net worth to 4,340, below leaving the game alone.

Tethers, 7 to 22 minutes. Luna kept adding ox tethers: 19 by 12:35 and 39 by 21:44, more than three for each quarry. Its reasoning gave two reasons: stone not yet arriving from quarries about 100 tiles from storage, and jobs, since each 5-wood tether employs a peasant and new peasants stop arriving while 24 stand idle ("Building more tethers will create additional jobs ... and help increase the population"). Later it saw "diminishing returns" at "about 2.58" tethers a quarry and built more anyway.

Short of gold, 17 to 33 minutes. Rations went to half at 17:43; at 19:27 it sold 20 pitch, dropped taxes to none and bought 200 more units of cheese and meat, which left 407 gold at 21 minutes. From then on it bought little, and the stone paid the bill: net worth climbed from 4,839 at 24 minutes to 7,354. In the last minute taxes went to extortionate (148 gold a month); popularity, which the score also counts, ended at 57 and falling.

Game minute	0	3	6	9	12	15	18	21	24	27	30	End
Net worth	4,561	5,951	5,363	5,207	4,521	4,340	4,953	4,685	4,839	5,238	6,493	7,354
Gold	4,340	4,667	3,735	3,562	1,988	1,557	1,711	407	381	402	935	1,246
Stone	0	50	50	58	82	130	202	250	282	346	442	530
Iron	0	15	23	31	40	48	56	66	70	78	90	100
Population	2	27	53	70	77	92	96	99	102	106	106	106
Popularity	100	100	94	100	95	90	77	68	68	74	60	57

What held it back

Food for 106 people, all bought. 630 units of cheese and meat cost 5,040 gold and 2,520 net worth against the sell price, and the run still ended on half rations. With wood, its purchases cost 3,675 net worth.
Twenty more tethers than the stone needed. After 15 minutes its stone came in at about 22 a minute, the same as Claude Opus 5.5's with 16 tethers for 10 quarries. The extra tethers added jobs and people rather than stone.
Spending ran ahead of income. Rations were halved at 17:43 and gold was down to about 400 from 21 to 27 minutes; popularity fell from 90 to 68 before the stone arrived.
Population stopped at 106 of 130 places from 27 minutes, with 24 idle at the end.

What worked

Twelve quarries and four iron mines by 12 minutes, with every tether beside the quarries: the most stone of the 1.5.6 runs (530).
The most people of the round. A chapel and an inn from 1:24, and a church from 13:21, kept popularity at 90 or more for the first 15 minutes.
Every reported placement matches the game's final count of its buildings.

Cost and speed

143 turns, 172 tool calls (7 waits), 6.60 million tokens and $0.125 (preparation $0.002, play $0.117, compaction $0.006), worked out from the list prices on the model profile. The interview cost $0.002 more, with 51,830 of its 54,942 tokens read from the cache. The agent run took 57.7 real minutes, 28.7 of them waiting for the model; the context was compacted three times (73 seconds).

Beside its run on 1.5.4 and the others on 1.5.6

	Luna, 1.5.6	Luna, 1.5.4	Claude Opus 5.5, 1.5.6	GLM 5.3 Flash, 1.5.6
Net worth	7,354	6,626	8,757	6,655
Gold / goods	1,246 / 6,108	3,437 / 3,189	2,513 / 6,244	4,560 / 2,095
Population / housing	106 / 130	78 / 162	58 / 58	70 / 98
Popularity	57	97	59	66
Quarries / ox tethers / iron mines	12 / 39 / 4	11 / 11 / 4	10 / 16 / 4	7 / 17 / 4
Stone / iron at the end	530 / 100	116 / 49	506 / 99	148 / 17
Net worth lost to purchases	3,675	4,500	2,640	2,890
Cost	$0.125		$2.012	$0.107

Luna and Opus built the same economy, quarries on the western stone and mines on the eastern ore, and dug the same stone and iron. Opus held its town at 58 people and kept twice the gold; Luna grew to 106 and paid for their food. Against its 1.5.4 run, the stone is the difference: 530 against 116 from about the same quarries. Luna put its tethers beside the quarries both times, as 1.5.6 now says to, but this time built 19 of them by 12 minutes, against 11 in all on 1.5.4.

The interview

Luna said it "concentrated on stone and iron production and treated the lack of farmland as a reason not to build a food-production economy", missing that bought wheat could feed a mill-and-bakery chain, and that it "spent too much attention on adding ox tethers instead of testing those higher-impact strategies". Playing again, it would test the wheat-to-bread chain early, build housing toward 250 with more inns, religion and fear-factor buildings, and "add mines and quarries only after checking that they are productive". It called the tools and readings useful and its strategy, not the tools, the problem.

For the harness

No placement failed for a harness reason. Four place_near calls found no room left against the keep (no_free_spot_in_view), and each time Luna placed the building by tile instead.
The game saved correctly at the end.

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	7,354	9,984
Gold	1,246	8,574
Stored goods, at the sell price	6,108	1,410
Population	106	250
Housing	130	250
Popularity	57	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?

GPT-6 Luna

I beat leaving the game alone, ending at 7,354 net worth versus 5,192, but I fell well short of the person’s 9,984. My population was 106 versus 250, and my popularity was 57 versus 100. I also ended with far more value sitting in goods and much less in gold.

The main difference was that I concentrated on stone and iron production and treated the lack of farmland as a reason not to build a food-production economy. I missed that imported wheat could support a large mill-and-bakery chain. I also built only 130 housing and one inn, while the person’s larger population, broad ale coverage, religion, and fear-factor buildings supported much greater growth and worker efficiency. I spent too much attention on adding ox tethers instead of testing those higher-impact strategies.

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?

GPT-6 Luna

I’d test the wheat-to-flour-to-bread chain early, even without farms: buy wheat, build a small mill-and-bakery setup, and check its actual output and marketplace returns before scaling it up. The person’s results show that this could support a much larger economy than I built.

I’d also expand housing toward 250, build services that sustain popularity at that scale—including more inns and religious buildings—and use fear-factor buildings to improve worker efficiency. I’d keep stone and iron production, but add mines and quarries only after checking that they are productive and adequately supplied with tethers. Finally, I’d track gold and stored goods more deliberately and avoid spending time on production that isn’t improving the overall economy.

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?

GPT-6 Luna

The tools and game readings were useful: the inventory and status data made it possible to track production, popularity, storage, and population, while find_sites and tile-based building placement helped with terrain-bound industries. The time budget made it important to build and adjust in batches, but I did not make the best use of it.

What got in my way was my strategy, not a major tool failure. I over-weighted raw-material production and failed to experiment early with imported inputs and processed goods. A clearer early indication of how much wheat a mill-and-bakery chain consumes and produces might have prompted me to test it sooner, though I should have tested it with the information available.

4Anything else

Interviewer

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

GPT-6 Luna

I’m glad the run beat the idle baseline, but the comparison makes the missed opportunity clear: I preserved a stable, productive mining settlement, yet didn’t turn it into a large, high-population economy.

Record

Attempts and Reproducibility

Every attempt1

Episode	Attempt	Outcome	Net worth	Counted
1	1	Full-budget score	7,354	Counted

What ran

Benchmark version: 1.5.6
Fingerprint: 3e1f84510204
Harness commit: 75acfb7
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 20261012T123257-cf634001