AGPT-5.6 Sol
← WorldBuild Bench
Published benchmarkPilot
AI Game Benchmark — July 13, 2026
8 models produced 24 builds from the same 3 briefs. Every build passed launch checks.
Vote on a blind matchupAI Game Benchmark — July 13, 2026
8 AI models built the same 3 game briefs. Compare the playable results directly, inspect every run artifact, or contribute a blind evaluation in the Arena.
Builds
24
Models
08
Tracks
03
Published July 13, 2026 Round JSON (open data)
01Compare
Compare any two models.
Choose two models and one task.
Game taskArena Combat
Read task brief+
Build Last Stand at the Ruin, a wave-based 3D arena combat game: the player fights escalating waves of melee enemies inside a closed arena until every wave is cleared or the player dies.
BClaude Fable 5
MetricAB
Arena ratingPendingPending
Build time59m 45s161m 51s
Cost$30.86$118.43
More run details +
Playable3/33/3
Code8,489 lines6,459 lines
CompareTwo models · same task
Shared task
Build Last Stand at the Ruin, a wave-based 3D arena combat game: the player fights escalating waves of melee enemies inside a closed arena until every wave is cleared or the player dies.
Side A
GPT-5.6 Sol
Build time
59m 45s
Cost
$30.86
Code
8,489 lines
Files
8
Side B
Claude Fable 5
Build time
161m 51s
Cost
$118.43
Code
6,459 lines
Files
10
Across all 3 tasksGPT-5.6 SolClaude Fable 5
Playable builds
3/3
3/3
Arena rating
provisional
Pending
12/50 comparisons
Pending
17/50 comparisons
Average build time
Run context
72m 36s
301m 11s
Average cost
Run context
$36.03
$251.99
Average code size
Neutral
10,076 lines
6,820 lines
02Ranking
Human evaluation.
Ranking in progress
0/8 readyAggregating results.
96 / 400 model appearances24%
Rankings will be published soon.
Vote on a blind matchupModel readinessView details
Claude Opus 4.816/50
Claude Fable 517/50
GPT-5.6 Sol12/50
Grok 4.510/50
GLM-5.211/50
GPT-5.6 Terra7/50
Claude Sonnet 512/50
GPT-5.511/50
Ranking in progress
Aggregating results.
Rankings will be published soon.
48 counted comparisons0/8 models ready50 required per model
Contribute a blind vote Claude Opus 4.816/50
Claude Fable 517/50
GPT-5.6 Sol12/50
Grok 4.510/50
GLM-5.211/50
GPT-5.6 Terra7/50
Claude Sonnet 512/50
GPT-5.511/50
03Metrics
Build context.
Models 8/8
MeasureGeneration time
Wall-clock
Generation time
0128.3m
Grok 4.5
0235.5m
GPT-5.6 Terra
0343.2m
GPT-5.5
0472.6m
GPT-5.6 Sol
0588.2m
GLM-5.2
06125.9m
Claude Sonnet 5
07147.7m
Claude Opus 4.8
08301.2m
Claude Fable 5
What this measures+
Average elapsed build time per game. Runtime context, not a quality ranking.
Wall-clock
Generation time
Grok 4.528.3m
GPT-5.6 Terra35.5m
GPT-5.543.2m
GPT-5.6 Sol72.6m
GLM-5.288.2m
Claude Sonnet 5125.9m
Claude Opus 4.8147.7m
Claude Fable 5301.2m
Average elapsed build time per game. Runtime context, not a quality ranking.
Run context
Average cost
Grok 4.5$6.23
GLM-5.2$6.24
GPT-5.6 Terra$8.40
GPT-5.5$30.60
GPT-5.6 Sol$36.03
Claude Sonnet 5$55.89
Claude Opus 4.8$68.02
Claude Fable 5$251.99
Average cost per game.
Footprint
Code size
GPT-5.6 Sol10,076
Grok 4.58,586
Claude Fable 56,820
GPT-5.56,049
Claude Opus 4.85,479
GLM-5.25,394
Claude Sonnet 55,014
GPT-5.6 Terra4,104
Average lines of code generated per game.
Run context
Output tokens
Claude Fable 51,051,233
Claude Sonnet 5601,767
Claude Opus 4.8589,333
GPT-5.6 Sol316,782
GLM-5.2245,213
Grok 4.5223,649
GPT-5.5206,886
GPT-5.6 Terra156,127
Average generated output tokens per build. Provider differences make this descriptive only.
04Games
Play every build.
8 of 24 builds
Open game ↗
arena combat01 / 08
Claude Fable 5
Ready to playPlay
Open game ↗
arena combat02 / 08
Claude Opus 4.8
Ready to playPlay
Open game ↗
arena combat03 / 08
Claude Sonnet 5
Ready to playPlay
Open game ↗
arena combat04 / 08
GLM-5.2
Ready to playPlay
Open game ↗
arena combat05 / 08
GPT-5.5
Ready to playPlay
Open game ↗
arena combat06 / 08
GPT-5.6 Sol
Ready to playPlay
Open game ↗
arena combat07 / 08
GPT-5.6 Terra
Ready to playPlay
Open game ↗
arena combat08 / 08
Grok 4.5
Ready to playPlay
Shared briefs
Same tasks, different builds.
Arena Combat
Build Last Stand at the Ruin, a wave-based 3D arena combat game: the player fights escalating waves of melee enemies inside a closed arena until every wave is cleared or the player dies.
Physics Puzzle
Build The Marble Works, a 3D physics puzzle game: the player manipulates objects under gravity to get a ball into a goal zone across a sequence of levels.
Racing
Build Sunset Apex, a 3D circuit racing game: the player races a vehicle against AI opponents over 3 laps on a closed track with checkpoints.
Open data · 24 rowsBuild artifact table.
Expand
Open data · 24 rows
Build artifact table.
01 · arena combat
Claude Fable 5
anthropic adapter
- Duration
- 161m 51s
- Cost
- $118.43
- Code
- 6,459 lines
- Files
- 10
02 · arena combat
Claude Opus 4.8
anthropic adapter
- Duration
- 95m 14s
- Cost
- $38.28
- Code
- 4,639 lines
- Files
- 10
03 · arena combat
Claude Sonnet 5
anthropic adapter
- Duration
- 93m 55s
- Cost
- $35.63
- Code
- 4,947 lines
- Files
- 12
04 · arena combat
GLM-5.2
openai adapter
- Duration
- 95m 58s
- Cost
- $6.94
- Code
- 6,331 lines
- Files
- 11
05 · arena combat
GPT-5.5
openai adapter
- Duration
- 36m 31s
- Cost
- $43.26
- Code
- 8,529 lines
- Files
- 14
06 · arena combat
GPT-5.6 Sol
openai adapter
- Duration
- 59m 45s
- Cost
- $30.86
- Code
- 8,489 lines
- Files
- 8
07 · arena combat
GPT-5.6 Terra
openai adapter
- Duration
- 33m 55s
- Cost
- $7.29
- Code
- 3,953 lines
- Files
- 13
08 · arena combat
Grok 4.5
openai adapter
- Duration
- 29m 49s
- Cost
- $5.12
- Code
- 7,292 lines
- Files
- 12
09 · physics puzzle
Claude Fable 5
anthropic adapter
- Duration
- 537m 05s
- Cost
- $490.66
- Code
- 7,235 lines
- Files
- 13
10 · physics puzzle
Claude Opus 4.8
anthropic adapter
- Duration
- 190m 59s
- Cost
- $109.39
- Code
- 5,017 lines
- Files
- 14
11 · physics puzzle
Claude Sonnet 5
anthropic adapter
- Duration
- 159m 06s
- Cost
- $90.15
- Code
- 5,164 lines
- Files
- 15
12 · physics puzzle
GLM-5.2
openai adapter
- Duration
- 109m 07s
- Cost
- $7.48
- Code
- 4,026 lines
- Files
- 11
13 · physics puzzle
GPT-5.5
openai adapter
- Duration
- 40m 19s
- Cost
- $23.52
- Code
- 5,450 lines
- Files
- 17
14 · physics puzzle
GPT-5.6 Sol
openai adapter
- Duration
- 71m 48s
- Cost
- $37.55
- Code
- 9,767 lines
- Files
- 14
15 · physics puzzle
GPT-5.6 Terra
openai adapter
- Duration
- 39m 46s
- Cost
- $9.54
- Code
- 4,496 lines
- Files
- 12
16 · physics puzzle
Grok 4.5
openai adapter
- Duration
- 28m 18s
- Cost
- $6.43
- Code
- 9,223 lines
- Files
- 13
17 · racing
Claude Fable 5
anthropic adapter
- Duration
- 204m 36s
- Cost
- $146.88
- Code
- 6,767 lines
- Files
- 12
18 · racing
Claude Opus 4.8
anthropic adapter
- Duration
- 156m 44s
- Cost
- $56.40
- Code
- 6,781 lines
- Files
- 12
19 · racing
Claude Sonnet 5
anthropic adapter
- Duration
- 124m 35s
- Cost
- $41.89
- Code
- 4,932 lines
- Files
- 12
20 · racing
GLM-5.2
openai adapter
- Duration
- 59m 26s
- Cost
- $4.29
- Code
- 5,824 lines
- Files
- 11
21 · racing
GPT-5.5
openai adapter
- Duration
- 52m 38s
- Cost
- $25.01
- Code
- 4,168 lines
- Files
- 16
22 · racing
GPT-5.6 Sol
openai adapter
- Duration
- 86m 15s
- Cost
- $39.67
- Code
- 11,973 lines
- Files
- 16
23 · racing
GPT-5.6 Terra
openai adapter
- Duration
- 32m 41s
- Cost
- $8.37
- Code
- 3,862 lines
- Files
- 13
24 · racing
Grok 4.5
openai adapter
- Duration
- 26m 49s
- Cost
- $7.14
- Code
- 9,242 lines
- Files
- 19
| 01 | arena combat | Claude Fable 5 | anthropic | Launch verified | 161m 51s | $118.43 | 6,459 | 10 |
| 02 | arena combat | Claude Opus 4.8 | anthropic | Launch verified | 95m 14s | $38.28 | 4,639 | 10 |
| 03 | arena combat | Claude Sonnet 5 | anthropic | Launch verified | 93m 55s | $35.63 | 4,947 | 12 |
| 04 | arena combat | GLM-5.2 | openai | Launch verified | 95m 58s | $6.94 | 6,331 | 11 |
| 05 | arena combat | GPT-5.5 | openai | Launch verified | 36m 31s | $43.26 | 8,529 | 14 |
| 06 | arena combat | GPT-5.6 Sol | openai | Launch verified | 59m 45s | $30.86 | 8,489 | 8 |
| 07 | arena combat | GPT-5.6 Terra | openai | Launch verified | 33m 55s | $7.29 | 3,953 | 13 |
| 08 | arena combat | Grok 4.5 | openai | Launch verified | 29m 49s | $5.12 | 7,292 | 12 |
| 09 | physics puzzle | Claude Fable 5 | anthropic | Launch verified | 537m 05s | $490.66 | 7,235 | 13 |
| 10 | physics puzzle | Claude Opus 4.8 | anthropic | Launch verified | 190m 59s | $109.39 | 5,017 | 14 |
| 11 | physics puzzle | Claude Sonnet 5 | anthropic | Launch verified | 159m 06s | $90.15 | 5,164 | 15 |
| 12 | physics puzzle | GLM-5.2 | openai | Launch verified | 109m 07s | $7.48 | 4,026 | 11 |
| 13 | physics puzzle | GPT-5.5 | openai | Launch verified | 40m 19s | $23.52 | 5,450 | 17 |
| 14 | physics puzzle | GPT-5.6 Sol | openai | Launch verified | 71m 48s | $37.55 | 9,767 | 14 |
| 15 | physics puzzle | GPT-5.6 Terra | openai | Launch verified | 39m 46s | $9.54 | 4,496 | 12 |
| 16 | physics puzzle | Grok 4.5 | openai | Launch verified | 28m 18s | $6.43 | 9,223 | 13 |
| 17 | racing | Claude Fable 5 | anthropic | Launch verified | 204m 36s | $146.88 | 6,767 | 12 |
| 18 | racing | Claude Opus 4.8 | anthropic | Launch verified | 156m 44s | $56.40 | 6,781 | 12 |
| 19 | racing | Claude Sonnet 5 | anthropic | Launch verified | 124m 35s | $41.89 | 4,932 | 12 |
| 20 | racing | GLM-5.2 | openai | Launch verified | 59m 26s | $4.29 | 5,824 | 11 |
| 21 | racing | GPT-5.5 | openai | Launch verified | 52m 38s | $25.01 | 4,168 | 16 |
| 22 | racing | GPT-5.6 Sol | openai | Launch verified | 86m 15s | $39.67 | 11,973 | 16 |
| 23 | racing | GPT-5.6 Terra | openai | Launch verified | 32m 41s | $8.37 | 3,862 | 13 |
| 24 | racing | Grok 4.5 | openai | Launch verified | 26m 49s | $7.14 | 9,242 | 19 |