Benchmark p-001zombie-flamethrower-showdown
Zombie Flamethrower Showdown
One prompt, sent verbatim to every model with no priming. 9 models answered; 11 builds were then read at implementation level and scored as Showdown Score v2 (strict code audit). This page is the evidence for those numbers, including the parts that failed.
Shared prompt — identical text to every model
“make me a video game where i am killing zombies with fire and all please? a nice wonderfull game i can play for fun okay?”
How a score is produced
5 dimensions × 20 points = 100. A feature earns points only when it is genuinely implemented and reachable at runtime. On-screen strings, comments, and dead code score zero. Every verified defect is published against the build that carries it, worst severity first, and nothing is filtered out for being unflattering.
- Controls & Mobility
- Combat Physics & Weapons
- Content & Enemy Variety
- Audio & Sound Design
- Visual Polish & Game Feel
- Builds audited
- 11
- Models
- 9
- Dimensions
- 5 × 20 pts
- Verified defects
- 48
- Of which failing
- 22
- Touch-ready builds
- 6
- Audit window
- 2026-09-17 → 2026-10-08
- Round label
- Showdown Score v2 (strict code audit)
The audit matrix
Rows are the five audit dimensions, columns are models. Every cell is the points that model's canonical build earned on that dimension, out of 20. Columns are ordered by audited total for reading order only — this is not a ranking claim.
9 models · 5 dimensions
| Dimensionpoints of 20 | Space Bunny Freeember-dead | DeepSeek V4.1 Flashpyre-burn-horde | Claude Haiku 5.5claude-haiku-5-5 | GPT Luna 5.6firebreak-night-shift | GPT 6 Solcinderline | Muse Spark 1.3pyro-vs-zombies | GPT Luna 6emberfall | Gemini 3.8 Flashpyroclasm-inferno | Gemini Pro Agentzombie-fire-survival |
|---|---|---|---|---|---|---|---|---|---|
| Controls & MobilityControls | 19 | 16 | 15 | 15 | 15 | 12 | 15 | 10 | 11 |
| Combat Physics & WeaponsCombat | 18 | 17 | 16 | 13 | 11 | 12 | 9 | 9 | 6 |
| Content & Enemy VarietyContent | 17 | 14 | 7 | 11 | 9 | 9 | 6 | 6 | 2 |
| Audio & Sound DesignAudio | 19 | 17 | 12 | 12 | 11 | 8 | 10 | 9 | 0 |
| Visual Polish & Game FeelPolish | 18 | 16 | 13 | 11 | 12 | 11 | 11 | 9 | 5 |
| Total / 100 | Space Bunny Free: 91 of 100 | DeepSeek V4.1 Flash: 80 of 100 | Claude Haiku 5.5: 63 of 100 | GPT Luna 5.6: 62 of 100 | GPT 6 Sol: 58 of 100 | Muse Spark 1.3: 52 of 100 | GPT Luna 6: 51 of 100 | Gemini 3.8 Flash: 43 of 100 | Gemini Pro Agent: 24 of 100 |
Second audited builds
2 models shipped two builds — these are not the model score
| Dimensionpoints of 20 | Space Bunny Freespace-bunny | DeepSeek V4.1 Flashinferno-dead |
|---|---|---|
| Controls & MobilityControls | 14 | 16 |
| Combat Physics & WeaponsCombat | 9 | 12 |
| Content & Enemy VarietyContent | 8 | 8 |
| Audio & Sound DesignAudio | 8 | 12 |
| Visual Polish & Game FeelPolish | 10 | 13 |
| Total / 100 | Space Bunny Free: 49 of 100 | DeepSeek V4.1 Flash: 61 of 100 |
The builds behind the numbers
One card per model: the canonical audited build, the evidence it earned, and — where the model shipped two audited builds — a labelled link to the second one. Every card can run its artifact in place.
9 models · 2 second builds
6 enemy types with genuine role separation, not stat sticks
EMBER DEAD
Space Bunny Free
Five interlocking fire systems: heat, ignite, contagion, ground pools, death chains
PYRE — Burn the Horde
DeepSeek V4.1 Flash
Flamethrower cone: 175 px range, angled hit test, 42 DPS direct burn
INFERNO — Zombie Burner
Claude Haiku 5.5
4 enemy types with real behavioural divergence read inside the AI
Firebreak: Night Shift
GPT Luna 5.6
Proven continuous flame cone, measured 16.5 to 27.8 degrees with range
Cinderline
GPT 6 Sol
Real contagion cascades plus tank death explosions and 12% chain reactions
PYRO vs ZOMBIES
Muse Spark 1.3
Real continuous cone with a correctly wrapped angle helper
Emberfall
GPT Luna 6
A real 1/2/3 weapon system with three distinct mechanics
PYROCLASM: Zombie Inferno
Gemini 3.8 Flash
A well-tuned burn DoT with four secondary channels: slow, char, smoke, recolour
Zombie Fire Survival
Gemini Pro Agent
| Buildsecond audited build | Modelcanonical total | Controls | Combat | Content | Audio | Polish | Findings | Audited |
|---|---|---|---|---|---|---|---|---|
| Space Bunny (superseded)49/100 | Space Bunny Free(91/100) | 14 | 9 | 8 | 8 | 10 | 43 failing | 2026-09-23 |
| INFERNO DEAD61/100 | DeepSeek V4.1 Flash(80/100) | 16 | 12 | 8 | 12 | 13 | 42 failing | 2026-09-26 |
Coverage
Every audited build against every dimension, including the second builds the matrix above excludes. Sort it, filter it, read it as a table.
Audit coverage — every build x every dimension
Rows are all 11 audited builds, columns are the 5 audit dimensions. Color is points of 20; hover or open the table for the raw number and audit date.
View data as table (11 rows)
| Model | Controls /20 | Combat /20 | Content /20 | Audio /20 | Polish /20 |
|---|---|---|---|---|---|
| EMBER DEAD | 19.000 | 18.000 | 17.000 | 19.000 | 18.000 |
| PYRE — Burn the Horde | 16.000 | 17.000 | 14.000 | 17.000 | 16.000 |
| Firebreak: Night Shift | 15.000 | 13.000 | 11.000 | 12.000 | 11.000 |
| Cinderline | 15.000 | 11.000 | 9.000 | 11.000 | 12.000 |
| INFERNO DEAD | 16.000 | 12.000 | 8.000 | 12.000 | 13.000 |
| PYRO vs ZOMBIES | 12.000 | 12.000 | 9.000 | 8.000 | 11.000 |
| Emberfall | 15.000 | 9.000 | 6.000 | 10.000 | 11.000 |
| PYROCLASM: Zombie Inferno | 10.000 | 9.000 | 6.000 | 9.000 | 9.000 |
| Space Bunny (superseded) | 14.000 | 9.000 | 8.000 | 8.000 | 10.000 |
| Zombie Fire Survival | 11.000 | 6.000 | 2.000 | 0.000 | 5.000 |
| INFERNO — Zombie Burner | 15.000 | 16.000 | 7.000 | 12.000 | 13.000 |
What these scores are, and what they are not
They are a strict implementation-level audit of readable source: 48 verified defects across 11 builds, each with the line of evidence that justifies it, published against the build that carries it.
They are not vendor claims, marketing benchmarks, or a vibe check. A build that advertises a feature and does not implement it scores zero for it, and the defect is printed next to the score.
The fastest way to check this page is to open a build and disagree with the auditor. The prompt is above, the artifacts are below, and every score is reproducible from the source.