The Lab

AI Game Benchmarks

I hand coding models an old game of mine and a goal. They get one shot, no hints, no follow-up. Everything they build is scored by the same harness and playable below.

4Benchmarks
47Runs
46Playable
20Models
Three experiments

The benchmarks

See the brief each model received, then compare what the models built.

46 builds, ready now

Play the builds

Pick a model and jump straight into its game.

Voidbreaker built by claude-sonnet-5-5 Voidbreaker claude-sonnet-5-5 via claude Play now → Voidbreaker built by gemini-3.7-flash-high Voidbreaker gemini-3.7-flash-high via antigravity Play now → Voidbreaker built by gpt-6.1-sol Voidbreaker gpt-6.1-sol via codex Play now → Voidbreaker built by grok-4.6 Voidbreaker grok-4.6 via grok Play now → Voidbreaker built by claude-fable-5-1-1m Voidbreaker claude-fable-5-1-1m via claude Play now → Voidbreaker built by gpt-6-astra Voidbreaker gpt-6-astra via codex Play now → Voidbreaker built by claude-fable-5 Voidbreaker claude-fable-5 via claude Play now → Voidbreaker built by sonnet Voidbreaker sonnet via claude Play now → Voidbreaker built by stealth-union-alpha Voidbreaker stealth-union-alpha via openrouter Play now → Voidbreaker built by grok-4.7 Voidbreaker grok-4.7 via grok Play now → Voidbreaker built by gpt-6-sol Voidbreaker gpt-6-sol via codex Play now → Voidbreaker built by gpt-5.6-sol Voidbreaker gpt-5.6-sol via codex Play now → Voidbreaker built by gpt-5.6-sol Voidbreaker gpt-5.6-sol via codex Play now → Voidbreaker built by claude-opus-5-5-1m Voidbreaker claude-opus-5-5-1m via claude Play now → Voidbreaker built by z-ai-glm-5.3-flash Voidbreaker z-ai-glm-5.3-flash via openrouter Play now → Voidbreaker built by gpt-5.6-terra Voidbreaker gpt-5.6-terra via codex Play now → Cruiser Reloaded built by gemini-3.7-flash-high Cruiser Reloaded gemini-3.7-flash-high via antigravity Play now → Cruiser Reloaded built by claude-sonnet-5-5 Cruiser Reloaded claude-sonnet-5-5 via claude Play now → Cruiser Reloaded built by gpt-6.1-sol Cruiser Reloaded gpt-6.1-sol via codex Play now → Cruiser Reloaded built by claude-fable-5 Cruiser Reloaded claude-fable-5 via claude Play now → Cruiser Reloaded built by gpt-6-astra Cruiser Reloaded gpt-6-astra via codex Play now → Cruiser Reloaded built by grok-4.6 Cruiser Reloaded grok-4.6 via grok Play now → Cruiser Reloaded built by grok-4.7 Cruiser Reloaded grok-4.7 via grok Play now → Cruiser Reloaded built by claude-fable-5-1-1m Cruiser Reloaded claude-fable-5-1-1m via claude Play now → Cruiser Reloaded built by claude-opus-5 Cruiser Reloaded claude-opus-5 via claude Play now → Cruiser Reloaded built by stealth-union-alpha Cruiser Reloaded stealth-union-alpha via openrouter Play now → Cruiser Reloaded built by claude-fable-5 Cruiser Reloaded claude-fable-5 via claude Play now → Cruiser Reloaded built by gpt-5.6-sol Cruiser Reloaded gpt-5.6-sol via codex Play now → Cruiser Reloaded built by gpt-5.6-terra Cruiser Reloaded gpt-5.6-terra via codex Play now → Cruiser Reloaded built by muse-spark-1.3-xhigh Cruiser Reloaded muse-spark-1.3-xhigh via cursor Play now → Cruiser Reloaded built by gpt-5.6-sol Cruiser Reloaded gpt-5.6-sol via codex Play now → Cruiser Reloaded built by gpt-6-sol Cruiser Reloaded gpt-6-sol via codex Play now → Cruiser Reloaded built by claude-opus-5-5-1m Cruiser Reloaded claude-opus-5-5-1m via claude Play now → Cruiser Reloaded built by z-ai-glm-5.3-flash Cruiser Reloaded z-ai-glm-5.3-flash via openrouter Play now → Retro Cruiser built by composer-2.5 Retro Cruiser composer-2.5 via cursor Play now → Retro Cruiser built by claude-fable-5 Retro Cruiser claude-fable-5 via claude Play now → Retro Cruiser built by grok-4.5-fast-xhigh Retro Cruiser grok-4.5-fast-xhigh via cursor Play now → Retro Cruiser built by gpt-5.5 Retro Cruiser gpt-5.5 via codex Play now → Void Cruiser 3D built by claude-sonnet-5-5 Void Cruiser 3D claude-sonnet-5-5 via claude Play now → Void Cruiser 3D built by gpt-6-astra Void Cruiser 3D gpt-6-astra via codex Play now → Void Cruiser 3D built by claude-opus-5-5-1m Void Cruiser 3D claude-opus-5-5-1m via claude Play now → Void Cruiser 3D built by gpt-6.1-sol Void Cruiser 3D gpt-6.1-sol via codex Play now → Void Cruiser 3D built by claude-fable-5-1-1m Void Cruiser 3D claude-fable-5-1-1m via claude Play now → Void Cruiser 3D built by gpt-6-sol Void Cruiser 3D gpt-6-sol via codex Play now → Void Cruiser 3D built by stealth-union-alpha Void Cruiser 3D stealth-union-alpha via openrouter Play now → Void Cruiser 3D built by gpt-6-astra Void Cruiser 3D gpt-6-astra via codex Play now →
Secondary evidence

Automated standings

Harness scores describe technical checks. They are useful for comparison, but they are not a final judgment of the game.

Retro CruiserVoidbreakerCruiser ReloadedVoid Cruiser 3D
gemini-3.7-flash-high antigravity — 82 view 5 min 98 view no time —
composer-2.5 cursor 93 view no time — — —
claude-sonnet-5-5 claude — 84 view 2 h 6 min 91 view 2 h 13 min 74 view 3 h 4 min
gpt-6.1-sol codex — 79 view 35 min 89 view 51 min 70 view 1 h 0 min
claude-fable-5 claude 75 view no time 75 view no time 86 view 53 min —
gpt-6-astra codex — 77 view 41 min 85 view 41 min 72 view 30 min
grok-4.6 grok — 78 view 21 min 84 view 19 min —
grok-4.7 grok — 67 view no time 83 view 49 min —
claude-fable-5-1-1m claude — 78 view 29 min 82 view 43 min 66 view 37 min
claude-opus-5 claude — — 79 view 1 h 16 min —
stealth-union-alpha openrouter — 73 view 22 min 78 view 34 min 63 view 29 min
grok-4.5-fast-xhigh cursor 75 view no time — — —
sonnet claude — 73 view no time 0 gate no time —
gpt-5.6-sol codex — 62 view 12 min 73 view 11 min —
gpt-5.6-terra codex — 52 view no time 72 view no time —
muse-spark-1.3-xhigh cursor — — 72 view 9 min —
claude-opus-5-5-1m claude — 56 view 25 min 61 view 32 min 71 view 1 h 4 min
gpt-6-sol codex — 67 view 20 min 70 view 16 min 65 view 19 min
gpt-5.5 codex 56 view no time — — —
z-ai-glm-5.3-flash openrouter — 55 view 12 min 47 view 16 min —
75–100 strong 50–74 partial 1–49 thin 0 gate failed not attempted run time (full bar is the slowest run)