One episode · Large Barren Desert

GLM 5.3 Flash

Benchmark 1.6.3Z.aiReasoning mediumPlayed by the maintainersFull-budget score

All of GLM 5.3 Flash’s results and comparisons

Not part of the evaluation: the game is published, but not ranked.

Score

12.6

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

Net worth

5,275

+83 over doing nothing, 2% of the person’s gain of 4,792

Population

24% of the person’s gain of 246 · popularity 89

Cost

$0.08

2.3M tokens

Time

33 game min

43 min of wall-clock time, 46 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

GLM 5.3 Flash
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

5,275net worth

Gold: 1,506
Stored goods, at sell price: 3,769
Over the baseline (do nothing): +83
Population / housing: 62 / 82
Popularity: 89
Food: 3
Structures (map-wide): 71
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 stock3,769 gold in goods

Good	Units	Price	Value
Gold	1,506	1	1,506
Stone	164	7	1,148
Ale	74	10	740
Iron	21	23	483
Swords	11	30	330
Bows	19	15	285
Metal armour	8	30	240
Spears	24	10	240
Pitch	20	10	200
Wood	91	1	91
Meat	3	4	12
Net worth			5,275

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

	This episode	The person	Left alone
Net worth	5,275	9,984	5,192
Gold	1,506	8,574
Stored goods, at the sell price	3,769	1,410
Population	62	250
Housing	82	250
Popularity	89	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: +3
Taxes: +1
Ale from the inn: +2
Taxes: No taxes, 0 gold a month
Rations: full; 1 of 1 foods eaten
Worker efficiency: 100%
Storage: 20 stockpile piles, 5 empty; 1 granary holding 3 food

Buildings45

Hovel9
Ox Tether7
Stockpile5
Iron Mine4
Quarry4
Woodcutter3
Armory2
Fletchers' Workshop2
Armorer's Workshop1
Blacksmith's Workshop1
Brewery1
Chapel1
Granary1
Inn1
Keep1
Marketplace1
Poleturner's Workshop1

Tool calls70

market_trade20
build_structure15
wait_and_observe8
find_sites7
set_tax7
expand_storage3
go_to_tile3
demolish_building2
set_rations2
center_on1
map_overview1
status1

Play

Turns: 46
Buildings placed / attempted: 38 / 43
Anchor-tool placements: 5
Tool errors: 2
Raw inputs sent to the game: 570
Context compactions: 0
Game time: 33 min

Compute

Tokens: 2.3M
Served from cache: 97%
Output tokens: 10K
Cost, as billed: $0.08
Wall-clock time: 43 min
of which inference: 16 min
Peak game memory: 3,487 MiB

The plan, before play

I understand: grow the economy of this Large Barren Desert free-build save for ~33 game minutes, judged on net worth first (gold + stored goods at sell prices), then population and popularity. No farms are possible (no oasis grass), so value comes from stone/iron production, weapons, taxes, and trading.

Opening plan:

Get a marketplace up early (it's also needed for trading).
Iron mines on the eastern ore strips (iron is worth 23/unit — highest value) + woodcutters.
Quarries + ox tethers on the western stone.
Housing (hovels) + keep popularity above 50, set modest taxes for gold income.
Weapon workshops (blacksmith/fletcher) + armory to convert cheap inputs into high-value goods; sell surplus.
Watch storage room; sell or expand before piles fill.

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: GLM 5.3 Flash (z-ai/glm-5.3-flash) on OpenRouter, providers z-ai/fp8 then siliconflow/fp8, reasoning medium, output limit 32,768 tokens
Benchmark: Large Barren Desert construction, version 1.6.3, from the save the human reference game 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 4e7470bc, series 20261014T070917-3e36f777, 2026-10-14: GLM's second game on this map, a new series with no memory of the first, played on the latest version (the first game, series 20261012T101643-b291d04d, played 1.5.6). The map's score for GLM is the mean of the two.

Result

	This game	First game	The person	Left alone
Net worth	5,275	6,655	9,984	5,192
Gold	1,506	4,560	8,574
Stored goods, at the sell price	3,769	2,095	1,410
Population / housing	62 / 82	70 / 98	250 / 250
Popularity	89 (+3 next month)	66	100

A full-budget score: the run used all 2,000 game seconds. Net worth grew by 83 over leaving the game alone, 2% of the person's 4,792, against 1,463 in the first game. The stored goods are 164 stone (1,148), 74 ale (740), 21 iron (483), 11 swords (330), 19 bows (285), 8 metal armour (240), 24 spears (240), 20 pitch (200), 91 wood (91) and 3 meat (12). The pitch and 20 of the ale came with the start.

How the game went

Building, 0 to 17 minutes. A marketplace and two hovels at 0:11, four stockpile piles by 1:39, then iron mines, quarries with ox tethers and woodcutters, and by 3:29 an armory, a fletcher's and a poleturner's workshop. An inn and a brewery followed at 7:11, a blacksmith's workshop at 9:09, two more quarries at 10:58, an armorer's and a second fletcher's workshop at 14:42 and a chapel at 16:55: 9 hovels, 7 tethers, 4 iron mines, 4 quarries, 3 woodcutters (a fourth was demolished) and five weapon workshops. The woodcutters went by the groves map_overview listed, two of them 140 tiles away.

Buying to keep going, 4 to 33 minutes. Meat every two or three minutes from 4:49, 525 in all, planks four times (300), and ale twice (65). Food ran out at about 27 minutes and popularity fell to 40 before the last purchases brought it back. Taxes moved seven times between none and high. The population reached 62 at 18 minutes and stayed there with 24 idle and 20 places free.

Game minute	0	3	6	9	12	15	18	21	24	27	30	End
Net worth	4,696	6,026	5,672	5,300	4,725	3,798	3,411	3,023	3,058	3,514	3,354	5,275
Gold	4,430	5,022	4,652	3,956	3,414	2,199	1,528	1,205	1,134	2,391	1,986	1,506
Iron	0	15	19	27	34	42	46	51	56	16	17	21
Stone	0	50	50	58	24	48	72	48	44	92	116	164
Food	47	39	51	70	76	74	59	58	52	0	17	3
Population	2	28	44	48	55	59	62	61	62	61	61	62
Popularity	100	95	66	60	75	76	61	52	52	40	66	89

Up to 30 minutes the net worth row counts gold, wood, stone, iron and food (at 4) as the recording logs them; weapons, ale and pitch are left out, so it reads low.

Beside its first game

	This game (1.6.3)	First game (1.5.6)
Net worth	5,275	6,655
Buildings	9 hovels, 4 iron mines, 4 quarries, 7 tethers, 5 weapon workshops, inn, brewery, chapel	11 hovels, 4 iron mines, 7 quarries, 17 tethers
Population / popularity	62 / 89	70 / 66
Bought	525 meat, 300 planks, 65 ale	meat, cheese, apples and wood
Net worth lost buying	3,650	2,890
Cost	$0.08	$0.11

The first game kept its gold and sold its stone and iron. This one ended with about 3,000 less gold, spent on food, planks for the workshops and ale, and with about 1,100 in weapons and 23 more points of popularity.

What held it back

Goods bought at a loss. 525 meat for 4,200 gold (worth 2,100), 300 planks for 1,200 (worth 300) and 65 ale for 1,300 (worth 650): 3,650 net worth lost, more than all its growth.
Woodcutters with no trees. Four woodcutters went up between 1:02 and 6:39. Wood rose only when it bought planks; the woodcutters cut nothing in 33 minutes (see For the harness).
A brewery with no hops. No farm can be built on this map, and it bought no hops.
24 idle peasants. The population stopped at 62 from 18 minutes with 20 places free.
Fewer quarries and tethers than the first game: 4 quarries and 7 tethers against 7 and 17.

Market

525 meat, 300 planks and 65 ale bought for 6,700 gold; 150 stone and 50 iron sold for 2,200. The purchases lost 3,650 net worth against the sell price (meat 8 for 4, planks 4 for 1, ale 20 for 10); selling leaves net worth unchanged. These are units at the market's fixed prices. One meat purchase at 19:26 bought nothing; it bought meat again at 20:10.

Cost and speed

46 turns, 70 tool calls (8 waits), 2.30 million tokens and $0.083 billed by OpenRouter (preparation $0.003, play $0.079). The interview cost about $0.003 more, with 72,576 of its 74,110 tokens read from the cache. The agent run took 42.6 real minutes, 16.3 of them waiting for the model; the context was never compacted. All 46 requests read the prompt cache, 97.7% of the input tokens.

The interview

GLM called the result "a clear loss", barely above leaving the game alone. It said it treated food as something to buy and eat where the person ran a wheat, flour and bread chain, that it never used the fear factor, and that it held back on population because each peasant ate bought meat. Playing again it would buy wheat for mills and bakeries from the start, buy ale instead of brewing it, build bad things with a church to balance them, build many more ox tethers, skip woodcutters and grow hovels hard. It found the tools well designed, and said the map overview's tree groves misled it into building four woodcutters, that production rates are invisible, and that how purchases count against net worth is spread across the rules.

For the harness

A stale map probe on the game machine. Like every game of this batch on Oasis by the Sea and Large Barren Desert, this game ran a crusader_probe built from 1.5.0's sources. It was built on 2026-10-13 at 21:28 UTC as a 1.5.0 game started. The 1.6.3 sync then copied sources that matched by hash, and Ninja found nothing to rebuild. The probe's only change since 1.5.0 is 1.5.4's check that tells trees from cacti, so map_overview counted cacti as tree groves. find_sites counts trees from the reader's tile data and was right. Everything else this game played was 1.6.3, and the game stands as played. The desert has no trees, but map_overview listed five cactus groves (13 "trees" 16 tiles from the keep, 14 at 140 tiles). find_sites returned trees: 0 at the 13-tree grove's centre, and GLM's four woodcutters, placed by the listed groves, cut nothing.
The chapel built at 16:55 does not appear among the popularity factors at the end (tax +1, inn +2).
Its first build_structure call, with five placements, was refused (at most four), and one demolish of the far woodcutters missed the building's body; both answers named the fix.
The game saved correctly at the end ("glm 5-3 flash 1-6-3 4e7470").

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	5,275	9,984
Gold	1,506	8,574
Stored goods, at the sell price	3,769	1,410
Population	62	250
Housing	82	250
Popularity	89	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?

GLM 5.3 Flash

My 5,275 vs their 9,984, and population 62 vs 250 — a clear loss. My settlement barely beat "doing nothing" (5,192). The person's run exposed two things I got badly wrong:

I treated food as a cost to be bought and eaten; they treated it as a production chain. They bought wheat (4.6/unit) and turned it into flour and bread sold at 10/unit — a value-added loop that generates gold. I bought meat at 8/unit to feed peasants, which was a pure -240 net-worth purchase every couple of months. That single mistake probably accounts for most of the gap.
I never used fear-factor buildings. 21 bad things gave them 110% worker efficiency — everything (quarries, mines, workshops) produced faster for a small popularity cost, which they offset with religion and inns. I didn't even consider gallows/stocks because I was guarding popularity, and popularity with no production behind it is worthless.

Also they scaled hard: 13 quarries, 37 ox tethers, 30 hovels, 250 population. I hesitated to grow population because each peasant eats bought meat; they grew it because each peasant at 110% efficiency more than paid for itself.

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?

GLM 5.3 Flash

From the start I would:

Buy wheat immediately, build mills and bakeries, and sell bread — the core income engine on a map with no farms.
Buy ale rather than brewing it (they had 11 inns and no brewery), to get the inn popularity cheaply.
Build fear-factor buildings early and compensate with a church/cathedral for the religion bonus.
Spam ox tethers (37 vs my 6) — I learned late that quarry throughput is limited by oxen walking distance, and tethers are the fix.
Skip woodcutters entirely: on this map every "grove" is inedible cacti, so wood is marketplace-only. I wasted four woodcutters and much of the first 10 minutes discovering that.
Grow hovels aggressively; population is both a scored quantity and the workforce that makes the production scale pay.

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?

GLM 5.3 Flash

The tools are well designed — tile-based placement with automatic retry, the reader for inventory/popularity, and map_overview were all genuinely useful and saved a lot of clicking. What got in my way:

The map-overview "tree groves" list included cacti, and find_sites reported "trees: 0" only subtly; I burned real game time and four buildings on that. A clearer signal ("no wood-producing trees on this map") would have saved my opening.
Production rates are invisible: I could only infer throughput from inventory deltas over long waits, which made diagnosing the slow iron mines and oxen-based stone delivery guesswork.
The net-worth accounting of purchases (buy price vs sell-price valuation) only appears across scattered rules text; I internalized it too late. The first 8 minutes of my run were effectively net-worth negative.
The 33-minute budget is tight and unforgiving of early misreads — which is arguably the point.

4Anything else

Interviewer

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

GLM 5.3 Flash

My run was dominated by a defensive mindset — protecting popularity, feeding peasants, avoiding losses — when the winning play was offensive: build a conversion chain that creates value and scale it. The person's build also shows that "wasteful-looking" moves (selling stone to buy bread inputs, harsh fear factor) are fine when the loop is net positive. If I had internalized "net worth = value-added throughput × time" in minute one instead of minute twenty, the result would have looked very different.

Record

Attempts and Reproducibility

Every attempt1

Episode	Attempt	Outcome	Net worth	Counted
1	1	Full-budget score	5,275	Counted

What ran

Benchmark version: 1.6.3
Fingerprint: 96c4682877ca
Harness commit: 1d0c697
Prompt / tools SHA-256: 3bb4a16 / e41a140
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 20261014T070917-3e36f777