Leaderboards
Per category: the AI judge's provisional ranks within each scored prompt, then how every model and harness finished, with the median cost and wall time of the finished runs. Scores are never averaged across prompts, and every figure shows the runs behind it.
SVG
1922 results from 292 prompts, 1614 finished. Browse them
AI judge, per prompt
AI judge: the test runner (model claude-opus-5-5), scoring by visual review of the 1600 px render against the pre-registered rubric. Ranks order the recorded assessments within one prompt; equal totals share a rank and are provisional. Small differences have not been tested for repeatability. Only each cell's latest attempt is ranked; earlier attempts keep their pages. How results are scored
Pelican riding a bicycle (control)
13 scored / 33 results; 4 unfinished; 20 unscored (unfinished results are included in the unscored count)
29 finished results
- 1
Claude Opus 5.5Claude CodeClaude Code 2.1.280, max effort, via Anthropic9.5/10· Draft score — AI judge16m 05s$2.27 reportedOK - 1
GPT-6 AstraCodex CLICodex CLI 0.156.1, xhigh effort, via OpenAI9.5/10· Draft score — AI judge3m 11s$0.74 estimatedOK - 1
Grok 4.7Grok BuildGrok Build 1.0.41, xhigh effort, via xAI9.5/10· Draft score — AI judge30m 14s$2.98 reportedOK - 1
Claude Fable 5.1OpenCodeOpenCode 1.18.32, max effort, via Anthropic9.5/10· Draft score — AI judge12m 08s$3.18 reportedOK - 1
GLM-5.3OpenCodeOpenCode 1.18.32, max effort, via Fireworks9.5/10· Draft score — AI judge9m 06s$0.33 reportedOK - 6
Claude Opus 5.5OpenCodeOpenCode 1.18.32, max effort, via Anthropic9.0/10· Draft score — AI judge22m 32s$5.02 reportedOK - 6
GPT-6 AstraOpenCodeOpenCode 1.18.32, xhigh effort, via OpenAI9.0/10· Draft score — AI judge7m 08s$0.94 reportedOK - 6
Kimi K3OpenCodeOpenCode 1.18.32, max effort, via Moonshot9.0/10· Draft score — AI judge11m 47s$0.49 reportedOK - 9
Claude Fable 5.1Claude CodeClaude Code 2.1.280, max effort, via Anthropic8.5/10· Draft score — AI judge14m 40s$4.18 reportedOK - 9
Kimi K3Kimi Code CLIKimi Code CLI 2.1.0, max effort, via Moonshot8.5/10· Draft score — AI judge8m 22s$0.48 estimatedOK - 11
Grok 4.7OpenCodeOpenCode 1.18.32, xhigh effort, via xAI7.5/10· Draft score — AI judge3m 26s$1.48 reportedOK - 12
DeepSeek V4.1 Flashdshdsh 0.1.5-rc.3, max effort, via DeepSeek7.0/10· Draft score — AI judge7m 48s$0.075 estimatedOK - 13
DeepSeek V4.1 FlashOpenCodeOpenCode 1.18.32, max effort, via DeepSeek6.5/10· Draft score — AI judge2m 47s$0.021 reportedOK
Results without a quality rank
- —
Claude Sonnet 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicUnscored: not judged yet26m 48s$6.66 reportedOK - —
GLM-5.3Claude CodeClaude Code 2.1.280, max effort, via FireworksUnscored: not judged yet8m 10s$0.55 estimatedOK - —
GPT-6 SolCodex CLICodex CLI 0.156.1, max effort, via OpenAIUnscored: not judged yet1m 54s$0.21 estimatedOK - —Claude Opus 5.5Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5, max effort, via AnthropicStopped by us — not a model failureUnscored — run did not finish1h 27mnot recorded: the run's token counts were never reported, so there is no estimateStopped by us
- —Claude Opus 5.5Hermes Agent 0.21.5Hermes Agent 0.21.5, max effort, via AnthropicStopped by us — not a model failureUnscored — run did not finish1h 06mnot recorded: the run's token counts were never reported, so there is no estimateStopped by us
- —
Claude Opus 5.5Kilo Code CLI 7.7.9Kilo Code CLI 7.7.9, max effort, via AnthropicUnscored: not judged yet21m 50s$4.29 reportedOK - —
Claude Opus 5.5Oh My Pi (omp) 18.3.1Oh My Pi (omp) 18.3.1, max effort, via AnthropicUnscored: not judged yet22m 49s$4.39 reportedOK - —
Claude Opus 5.5Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills)Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills), max effort, via AnthropicUnscored: not judged yet10m 55s$2.03 reportedOK - —
Claude Opus 5.5Pi 0.87.1Pi 0.87.1, max effort, via AnthropicUnscored: not judged yet22m 40s$4.61 reportedOK - —
Claude Sonnet 5.5OpenCodeOpenCode 1.18.33, max effort, via AnthropicUnscored: not judged yet29m 02s$5.06 reportedOK - —
DeepSeek V4.1 FlashCline CLICline CLI 3.0.65, high effort, via DeepSeekUnscored: not judged yet13m 15snot recorded: the run's usage was never recordedOK - —
Gemini 3.8 FlashCline CLICline CLI 3.0.65, high effort, via GoogleUnscored: not judged yet6m 14s$0 reportedOK - —
Space Bunny AlphaCline CLICline CLI 3.0.65, high effortUnscored: not judged yet1m 26s$0 reportedOK - —
Space Bunny AlphaOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet16m 48s$0 reportedOK - —
Muse Spark 1.3Cline CLICline CLI 3.0.65, high effortUnscored: not judged yet3m 16s$0 reportedOK - —
Muse Spark 1.3OpenCodeOpenCode 1.18.32, xhigh effortUnscored: not judged yet2m 40s$0 reportedOK - —
LongCat 2.5 PreviewOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet9m 13s$0 reportedOK - —
MiMo V2.6 FlashCline CLICline CLI 3.0.65, high effortUnscored: not judged yet48m 17s$0 reportedOK - —Gemini 3.6 Flashraw API callRaw API call, high effort, via GoogleErrorUnscored — run did not finish4.0 s$0 estimatedError
- —Gemini 3.7 Flashraw API callRaw API call, high effort, via GoogleErrorUnscored — run did not finish5.5 snot recorded: the run's usage was never recordedError
MacBook Pro 16" realistic 3D render (Playcode benchmark)
12 scored / 25 results; 5 unfinished; 13 unscored (unfinished results are included in the unscored count)
Unranked — criterion breakdowns need repair
20 finished results
- —
Claude Opus 5.5Claude CodeClaude Code 2.1.280, max effort, via Anthropic9.5/10· Draft score — AI judge38m 51s$7.14 reportedOK - —
Claude Fable 5.1Claude CodeClaude Code 2.1.280, max effort, via Anthropic9.0/10· Draft score — AI judge25m 14s$8.70 reportedOK - —
GPT-6 AstraCodex CLICodex CLI 0.156.1, xhigh effort, via OpenAI8.5/10· Draft score — AI judge15m 48s$2.23 estimatedOK - —
DeepSeek V4.1 Flashdshdsh 0.1.5-rc.3, max effort, via DeepSeek7.0/10· Draft score — AI judge10m 40s$0.079 estimatedOK - —
Grok 4.7Grok BuildGrok Build 1.0.41, xhigh effort, via xAI8.5/10· Draft score — stored total; criterion breakdown unavailable41m 45s$7.66 reportedOK - —
Kimi K3Kimi Code CLIKimi Code CLI 2.1.0, max effort, via Moonshot8.0/10· Draft score — stored total; criterion breakdown unavailable27m 44s$1.68 estimatedOK - —
Claude Fable 5.1OpenCodeOpenCode 1.18.32, max effort, via Anthropic9.0/10· Draft score — AI judge33m 13s$10.27 reportedOK - —
GPT-6 AstraOpenCodeOpenCode 1.18.32, xhigh effort, via OpenAI9.0/10· Draft score — AI judge10m 11s$1.94 reportedOK - —
DeepSeek V4.1 FlashOpenCodeOpenCode 1.18.32, max effort, via DeepSeek6.5/10· Draft score — AI judge10m 29s$0.092 reportedOK - —
Grok 4.7OpenCodeOpenCode 1.18.32, xhigh effort, via xAI4.5/10· Draft score — stored total; criterion breakdown unavailable40m 31s$15.78 reportedOK - —
Kimi K3OpenCodeOpenCode 1.18.32, max effort, via Moonshot8.5/10· Draft score — stored total; criterion breakdown unavailable43m 15s$2.59 reportedOK - —
GLM-5.3OpenCodeOpenCode 1.18.32, max effort, via Fireworks9.0/10· Draft score — stored total; criterion breakdown unavailable32m 12s$2.40 reportedOK
Results without a quality rank
- —
Claude Sonnet 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicUnscored: not judged yet31m 08s$6.64 reportedOK - —
GLM-5.3Claude CodeClaude Code 2.1.280, max effort, via FireworksUnscored: not judged yet22m 06s$1.11 estimatedOK - —
GPT-6 SolCodex CLICodex CLI 0.156.1, max effort, via OpenAIUnscored: not judged yet3m 55s$0.43 estimatedOK - —
Claude Opus 5.5OpenCodeOpenCode 1.18.32, max effort, via AnthropicUnscored: not judged yet38m 55s$8.38 reportedOK - —Claude Sonnet 5.5OpenCodeOpenCode 1.18.33, max effort, via AnthropicStopped (our timeout) — not a model failureUnscored — run did not finish45m 57s$4.58 reportedStopped (our timeout)
- —DeepSeek V4.1 FlashCline CLICline CLI 3.0.65, high effort, via DeepSeekErrorUnscored — run did not finish23m 23snot recorded: the run's token counts were never reported, so there is no estimateError
- —Gemini 3.8 FlashCline CLICline CLI 3.0.65, high effort, via GoogleErrorUnscored — run did not finish7m 20snot recorded: the run's token counts were never reported, so there is no estimateError
- —
Space Bunny AlphaCline CLICline CLI 3.0.65, high effortUnscored: not judged yet2m 26s$0 reportedOK - —
Space Bunny AlphaOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet6m 50s$0 reportedOK - —
Muse Spark 1.3Cline CLICline CLI 3.0.65, high effortUnscored: not judged yet12m 37s$0 reportedOK - —
Muse Spark 1.3OpenCodeOpenCode 1.18.32, xhigh effortUnscored: not judged yet5m 38s$0 reportedOK - —MiMo V2.6 FlashCline CLICline CLI 3.0.65, high effortErrorUnscored — run did not finish22m 04snot recorded: the run's token counts were never reported, so there is no estimateError
- —Gemini 3.6 Flashraw API callRaw API call, high effort, via GoogleDNF — model did not finishUnscored — run did not finish9.8 s$0 estimatedDNF
Lighthouse on a headland at night
12 scored / 26 results; 6 unfinished; 14 unscored (unfinished results are included in the unscored count)
20 finished results
- 1
Claude Fable 5.1OpenCodeOpenCode 1.18.32, max effort, via Anthropic9.0/10· Draft score — AI judge25m 12s$8.05 reportedOK - 2
Claude Fable 5.1Claude CodeClaude Code 2.1.280, max effort, via Anthropic8.5/10· Draft score — AI judge33m 32s$12.21 reportedOK - 2
Grok 4.7Grok BuildGrok Build 1.0.41, xhigh effort, via xAI8.5/10· Draft score — AI judge36m 05s$5.42 reportedOK - 2
Claude Opus 5.5OpenCodeOpenCode 1.18.32, max effort, via Anthropic8.5/10· Draft score — AI judge1h 15m$16.76 reportedOK - 2
Kimi K3OpenCodeOpenCode 1.18.32, max effort, via Moonshot8.5/10· Draft score — AI judge25m 50s$1.12 reportedOK - 6
GPT-6 AstraCodex CLICodex CLI 0.156.1, xhigh effort, via OpenAI8.0/10· Draft score — AI judge9m 08s$2.28 estimatedOK - 6
DeepSeek V4.1 Flashdshdsh 0.1.5-rc.3, max effort, via DeepSeek8.0/10· Draft score — AI judge13m 40s$0.24 estimatedOK - 6
GPT-6 AstraOpenCodeOpenCode 1.18.32, xhigh effort, via OpenAI8.0/10· Draft score — AI judge18m 29s$2.95 reportedOK - 9
Kimi K3Kimi Code CLIKimi Code CLI 2.1.0, max effort, via Moonshot7.5/10· Draft score — AI judge21m 53s$1.28 estimatedOK - 9
Grok 4.7OpenCodeOpenCode 1.18.32, xhigh effort, via xAI7.5/10· Draft score — AI judge23m 31s$3.49 reportedOK - 11
DeepSeek V4.1 FlashOpenCodeOpenCode 1.18.32, max effort, via DeepSeek6.5/10· Draft score — AI judge12m 10s$0.081 reportedOK - 12
GLM-5.3OpenCodeOpenCode 1.18.32, max effort, via Fireworks5.5/10· Draft score — AI judge13m 48s$0.75 reportedOK
Results without a quality rank
- —
Claude Opus 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicUnscored: not judged yet1h 15m$12.02 reportedOK - —Claude Sonnet 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicStopped (our timeout) — not a model failureUnscored — run did not finish45m 59snot recorded: the run's token counts were never reported, so there is no estimateStopped (our timeout)
- —
GLM-5.3Claude CodeClaude Code 2.1.280, max effort, via FireworksUnscored: not judged yet16m 19s$1.49 estimatedOK - —
GPT-6 SolCodex CLICodex CLI 0.156.1, max effort, via OpenAIUnscored: not judged yet5m 35s$0.70 estimatedOK - —Claude Sonnet 5.5OpenCodeOpenCode 1.18.33, max effort, via AnthropicStopped (our timeout) — not a model failureUnscored — run did not finish45m 59s$7.60 reportedStopped (our timeout)
- —DeepSeek V4.1 FlashCline CLICline CLI 3.0.65, high effort, via DeepSeekStopped by us — not a model failureUnscored — run did not finish42m 25snot recorded: the run's token counts were never reported, so there is no estimateStopped by us
- —
Gemini 3.8 FlashCline CLICline CLI 3.0.65, high effort, via GoogleUnscored: not judged yet6m 46s$0 reportedOK - —
Space Bunny AlphaCline CLICline CLI 3.0.65, high effortUnscored: not judged yet2m 45s$0 reportedOK - —
Space Bunny AlphaOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet7m 29s$0 reportedOK - —
Muse Spark 1.3Cline CLICline CLI 3.0.65, high effortUnscored: not judged yet49m 16snot recorded: the run's usage was never recordedOK - —
Muse Spark 1.3OpenCodeOpenCode 1.18.32, xhigh effortUnscored: not judged yet7m 34s$0 reportedOK - —MiMo V2.6 FlashCline CLICline CLI 3.0.65, high effortStopped by us — not a model failureUnscored — run did not finish42m 40snot recorded: the run's token counts were never reported, so there is no estimateStopped by us
- —Gemini 3.6 Flashraw API callRaw API call, high effort, via GoogleDNF — model did not finishUnscored — run did not finish2.8 s$0 estimatedDNF
- —Gemini 3.7 Flashraw API callRaw API call, high effort, via GoogleDNF — model did not finishUnscored — run did not finish8.0 s$0 estimatedDNF
Isometric SVG town with sprite sheet
11 scored / 26 results; 5 unfinished; 15 unscored (unfinished results are included in the unscored count)
Unranked — criterion breakdowns need repair
21 finished results
- —
Claude Fable 5.1Claude CodeClaude Code 2.1.280, max effort, via Anthropic9.5/10· Draft score — stored total; criterion breakdown unavailable39m 42s$12.27 reportedOK - —
GPT-6 AstraCodex CLICodex CLI 0.156.1, xhigh effort, via OpenAI9.5/10· Draft score — stored total; criterion breakdown unavailable11m 58s$2.47 estimatedOK - —
DeepSeek V4.1 Flashdshdsh 0.1.5-rc.3, max effort, via DeepSeek9.5/10· Draft score — stored total; criterion breakdown unavailable16m 20s$0.32 estimatedOK - —
Grok 4.7Grok BuildGrok Build 1.0.41, xhigh effort, via xAI9.5/10· Draft score — AI judge40m 44s$3.59 reportedOK - —
Kimi K3Kimi Code CLIKimi Code CLI 2.1.0, max effort, via Moonshot9.0/10· Draft score — AI judge48m 56s$2.54 estimatedOK - —
Claude Fable 5.1OpenCodeOpenCode 1.18.32, max effort, via Anthropic10.0/10· Draft score — stored total; criterion breakdown unavailable39m 11s$13.67 reportedOK - —
GPT-6 AstraOpenCodeOpenCode 1.18.32, xhigh effort, via OpenAI9.5/10· Draft score — stored total; criterion breakdown unavailable13m 27s$2.65 reportedOK - —
DeepSeek V4.1 FlashOpenCodeOpenCode 1.18.32, max effort, via DeepSeek9.0/10· Draft score — stored total; criterion breakdown unavailable26m 14s$0.23 reportedOK - —
Grok 4.7OpenCodeOpenCode 1.18.32, xhigh effort, via xAI9.0/10· Draft score — AI judge47m 32s$9.37 reportedOK - —
Kimi K3OpenCodeOpenCode 1.18.32, max effort, via Moonshot9.0/10· Draft score — AI judge30m 09s$1.34 reportedOK - —
GLM-5.3OpenCodeOpenCode 1.18.32, max effort, via Fireworks9.5/10· Draft score — AI judge48m 32s$3.46 reportedOK
Results without a quality rank
- —
Claude Opus 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicUnscored: not judged yet1h 02m$11.82 reportedOK - —Claude Sonnet 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicErrorUnscored — run did not finish26m 32s$4.00 reportedError
- —
GLM-5.3Claude CodeClaude Code 2.1.280, max effort, via FireworksUnscored: not judged yet46m 16s$3.24 estimatedOK - —
GPT-6 SolCodex CLICodex CLI 0.156.1, max effort, via OpenAIUnscored: not judged yet4m 34s$0.54 estimatedOK - —
Claude Opus 5.5OpenCodeOpenCode 1.18.32, max effort, via AnthropicUnscored: not judged yet59m 06s$15.39 reportedOK - —Claude Sonnet 5.5OpenCodeOpenCode 1.18.33, max effort, via AnthropicStopped (our timeout) — not a model failureUnscored — run did not finish46m 05s$8.80 reportedStopped (our timeout)
- —DeepSeek V4.1 FlashCline CLICline CLI 3.0.65, high effort, via DeepSeekErrorUnscored — run did not finish33m 47snot recorded: the run's token counts were never reported, so there is no estimateError
- —
Gemini 3.8 FlashCline CLICline CLI 3.0.65, high effort, via GoogleUnscored: not judged yet34m 59snot recorded: the run's token counts were never reported, so there is no estimateOK - —
Space Bunny AlphaCline CLICline CLI 3.0.65, high effortUnscored: not judged yet2m 32s$0 reportedOK - —
Space Bunny AlphaOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet7m 30s$0 reportedOK - —
Muse Spark 1.3Cline CLICline CLI 3.0.65, high effortUnscored: not judged yet5m 20s$0 reportedOK - —
Muse Spark 1.3OpenCodeOpenCode 1.18.32, xhigh effortUnscored: not judged yet6m 40s$0 reportedOK - —
LongCat 2.5 PreviewOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet25m 37s$0 reportedOK - —MiMo V2.6 FlashCline CLICline CLI 3.0.65, high effortErrorUnscored — run did not finish29m 58snot recorded: the run's token counts were never reported, so there is no estimateError
- —Gemini 3.6 Flashraw API callRaw API call, high effort, via GoogleDNF — model did not finishUnscored — run did not finish4.0 s$0 estimatedDNF
Animated pelican riding a bicycle (SMIL/CSS)
12 scored / 30 results; 5 unfinished; 18 unscored (unfinished results are included in the unscored count)
25 finished results
- 1
Claude Opus 5.5Claude CodeClaude Code 2.1.280, max effort, via Anthropic10.0/10· Draft score — AI judge35m 10s$4.96 reportedOK - 2
Kimi K3Kimi Code CLIKimi Code CLI 2.1.0, max effort, via Moonshot9.5/10· Draft score — AI judge13m 32s$0.74 estimatedOK - 2
Claude Opus 5.5OpenCodeOpenCode 1.18.32, max effort, via Anthropic9.5/10· Draft score — AI judge31m 51s$6.23 reportedOK - 2
Claude Fable 5.1OpenCodeOpenCode 1.18.32, max effort, via Anthropic9.5/10· Draft score — AI judge15m 09s$4.60 reportedOK - 2
GPT-6 AstraOpenCodeOpenCode 1.18.32, xhigh effort, via OpenAI9.5/10· Draft score — AI judge8m 47s$1.61 reportedOK - 6
Claude Fable 5.1Claude CodeClaude Code 2.1.280, max effort, via Anthropic9.0/10· Draft score — AI judge15m 10s$4.65 reportedOK - 6
GPT-6 AstraCodex CLICodex CLI 0.156.1, xhigh effort, via OpenAI9.0/10· Draft score — AI judge5m 36s$1.22 estimatedOK - 6
DeepSeek V4.1 Flashdshdsh 0.1.5-rc.3, max effort, via DeepSeek9.0/10· Draft score — AI judge10m 40s$0.038 estimatedOK - 6
Kimi K3OpenCodeOpenCode 1.18.32, max effort, via Moonshot9.0/10· Draft score — AI judge32m 30s$1.28 reportedOK - 6
GLM-5.3OpenCodeOpenCode 1.18.32, max effort, via Fireworks9.0/10· Draft score — AI judge30m 50s$1.95 reportedOK - 11
Grok 4.7OpenCodeOpenCode 1.18.32, xhigh effort, via xAI8.5/10· Draft score — AI judge24m 29s$7.17 reportedOK - 12
DeepSeek V4.1 FlashOpenCodeOpenCode 1.18.32, max effort, via DeepSeek8.0/10· Draft score — AI judge12m 08s$0.067 reportedOK
Results without a quality rank
- —
Claude Sonnet 5.5Claude CodeClaude Code 2.1.283, max effort, via AnthropicUnscored: not judged yet35m 45s$7.86 reportedOK - —
GLM-5.3Claude CodeClaude Code 2.1.280, max effort, via FireworksUnscored: not judged yet42m 04s$1.83 estimatedOK - —
GPT-6 SolCodex CLICodex CLI 0.156.1, max effort, via OpenAIUnscored: not judged yet2m 25s$0.28 estimatedOK - —Claude Opus 5.5Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5, max effort, via AnthropicStopped by us — not a model failureUnscored — run did not finish42m 03snot recorded: the run's token counts were never reported, so there is no estimateStopped by us
- —Claude Opus 5.5Hermes Agent 0.21.5Hermes Agent 0.21.5, max effort, via AnthropicStopped by us — not a model failureUnscored — run did not finish42m 03snot recorded: the run's token counts were never reported, so there is no estimateStopped by us
- —
Claude Opus 5.5Kilo Code CLI 7.7.9Kilo Code CLI 7.7.9, max effort, via AnthropicUnscored: not judged yet27m 05s$5.71 reportedOK - —
Claude Opus 5.5Oh My Pi (omp) 18.3.1Oh My Pi (omp) 18.3.1, max effort, via AnthropicUnscored: not judged yet33m 55s$8.07 reportedOK - —
Claude Opus 5.5Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills)Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills), max effort, via AnthropicUnscored: not judged yet23m 06s$3.93 reportedOK - —
Claude Opus 5.5Pi 0.87.1Pi 0.87.1, max effort, via AnthropicUnscored: not judged yet31m 38s$4.89 reportedOK - —
Claude Sonnet 5.5OpenCodeOpenCode 1.18.33, max effort, via AnthropicUnscored: not judged yet39m 43s$7.61 reportedOK - —DeepSeek V4.1 FlashCline CLICline CLI 3.0.65, high effort, via DeepSeekErrorUnscored — run did not finish32m 32snot recorded: the run's token counts were never reported, so there is no estimateError
- —
Gemini 3.8 FlashCline CLICline CLI 3.0.65, high effort, via GoogleUnscored: not judged yet25m 30snot recorded: the run's token counts were never reported, so there is no estimateOK - —
Space Bunny AlphaCline CLICline CLI 3.0.65, high effortUnscored: not judged yet1m 46s$0 reportedOK - —
Space Bunny AlphaOpenCodeOpenCode 1.18.32, max effortUnscored: not judged yet6m 12s$0 reportedOK - —
Muse Spark 1.3Cline CLICline CLI 3.0.65, high effortUnscored: not judged yet7m 44s$0 reportedOK - —
Muse Spark 1.3OpenCodeOpenCode 1.18.32, xhigh effortUnscored: not judged yet5m 51s$0 reportedOK - —MiMo V2.6 FlashCline CLICline CLI 3.0.65, high effortErrorUnscored — run did not finish22m 05snot recorded: the run's token counts were never reported, so there is no estimateError
- —Gemini 3.6 Flashraw API callRaw API call, high effort, via GoogleDNF — model did not finishUnscored — run did not finish3.7 s$0 estimatedDNF
Unscored prompts (no published rubric yet, so no ranks): DualSense controller, max fidelity (32 results), California brown pelican riding a bicycle (detailed spec) (29 results), Xbox 360 controller (23 results), BMW M4 Competition, side view (flat vector) (24 results), DualSense controller, quick (2 results), PS5 Pro console scene (2 results), whale_clouds (20 results), dolphin_vlog_bicycle (24 results), pelican_car_france (26 results), Twin-Bell Alarm Clock (14 results), Alien Planet (14 results), Ant Colony Cutaway (14 results), Flat App Icon Set (14 results), Art Deco Skyscraper (14 results), Floating Astronaut (14 results), Stargazer with Telescope (14 results), Autumn Forest Path (14 results), Ballet Dancer (14 results), Barn Owl in Flight (14 results), Battery Level Icons (14 results), Breakfast Skillet (14 results), Bubble Tea (14 results), Exploded Burger (14 results), Calendar Month View (14 results), Canyon River (14 results), svg_cargo_ship_containers__agent (1 result), Container Ship (14 results), Chameleon on a Branch (14 results), Chat Conversation UI (14 results), Chef Flipping a Pan (14 results), Circuit Board Close-Up (14 results), Corgi Catching a Frisbee (14 results), Cyberpunk Alley (14 results), Sales Dashboard Card (14 results), Desert Dunes at Night (12 results), Desk Lamp and Books (11 results), Noir Detective (10 results), Group Selfie (10 results), Dragon Coiled on a Tower (10 results), Dragonfly Over Pond (10 results), Quadcopter Drone (10 results), Windmill by a Canal (10 results), Pyramids at Sunset (10 results), Elephant Calf Mud Bath (13 results), File Upload Dropzone (10 results), 35mm Film Camera (10 results), Flamingo on One Leg (10 results), Floating Islands (10 results), Fountain Pen and Ink (10 results), Fruit Bowl Still Life (10 results), Retro Handheld Console (10 results), Gardener Watering Plants (10 results), Giraffe Side Profile (13 results), Victorian Glasshouse (10 results), Good Vibes Retro Type (10 results), Gothic Cathedral Facade (10 results), Greek Temple (9 results), Over-Ear Headphones (7 results), Hedgehog Under Mushroom (7 results), Rescue Helicopter (7 results), Hello Neon Sign (7 results), Honeybee on Honeycomb (7 results), Hot Air Balloons (7 results), Hourglass (7 results), svg_hummingbird_flower__agent (1 result), Hummingbird at Flower (7 results), Melting Ice Cream Cone (7 results), Five-Storey Pagoda (7 results), Ring of Keys (7 results), Kids Building a Snowman (7 results), Kingfisher Diving (7 results), Knight in Armour (7 results), svg_ladybug_leaf__agent (1 result), Ladybug on a Leaf (7 results), Latte Art (7 results), Lavender Fields (5 results), Layer Cake Slice (4 results), Line Icon Set (4 results), Login Screen (4 results), Giant Mech (3 results), Mechanical Keyboard (2 results), Medieval Castle (2 results), Mermaid on a Rock (2 results), Monarch Butterfly (4 results), Monster Truck Jump (2 results), Domed Mosque (2 results), Mountain Lake at Dawn (2 results), Music Player UI (2 results), Northern Lights (2 results), Octopus Juggling Teacups (13 results), Old Fisherman (2 results), Painter at an Easel (2 results), svg_panda_bamboo__agent (1 result), Panda Eating Bamboo (11 results), Peacock Tail Display (4 results), Emperor Penguin Huddle (2 results), Phoenix Rising (2 results), Pizza Slice Cheese Pull (2 results), Polar Bear on Ice Floe (7 results), Potted Succulents (2 results), Fitness Activity Rings (2 results), Bowl of Ramen (2 results), Double-Decker Bus (2 results), svg_red_fox_snow__agent (1 result), Red Fox in Snow (13 results), Retro Desktop Computer (2 results), Rice Terraces (2 results), Robot Vacuum (2 results), Robot Tending a Flower (2 results), Colourful Row Houses (2 results), Sailboat Regatta (2 results), Swimming Sea Turtle (12 results), Server Rack (2 results), Vintage Sewing Machine (2 results), Skateboarder Ollie (2 results), Sleeping Curled Cat (14 results), Exploded Smartphone (2 results), Smartwatch (2 results), Running Shoe (2 results), Rocket Launch (2 results), Starfighter in a Hangar (2 results), Stack of Pancakes (2 results), svg_steam_locomotive__agent (1 result), Steam Locomotive (2 results), Steampunk Airship (2 results), Deep-Sea Submarine (2 results), Sushi Platter (2 results), Suspension Bridge in Fog (2 results), Multi-Tool Pocket Knife (2 results), Trio of Tacos (2 results), Prairie Thunderstorm (2 results), Settings Toggles (2 results), Open Toolbox (2 results), Toucan on a Branch (2 results), Cable Car on a Hill (2 results), Treehouse (2 results), Tropical Beach (2 results), Open Umbrella in Rain (2 results), Kelp Forest (2 results), Unicorn in a Meadow (2 results), Vintage Biplane (2 results), Vintage Scooter (2 results), Vintage Typewriter (2 results), Erupting Volcano (2 results), VR Headset (2 results), Jungle Waterfall (2 results), Weather Widget (2 results), Wizard Casting a Spell (2 results), Wolf Howling at Moon (9 results), Acoustic Guitar (14 results), Illustrated Alphabet A B C (14 results), svg_astronaut_floating__agent (1 result), Bohr Model of an Atom (14 results), Fruit Sales Bar Chart (14 results), Baseball Stadium (14 results), Basketball Slam Dunk (14 results), Bauhaus Composition (14 results), Labelled Bicycle Diagram (14 results), svg_boxing_ring__agent (1 result), Boxing Ring (14 results), Butterfly Life Cycle (14 results), Campus Map (14 results), svg_chat_interface__agent (1 result), Chessboard Mid-Game (14 results), svg_city_park_top_down__agent (1 result), Top-Down City Park (14 results), Chalkboard Coffee Lettering (14 results), svg_cyberpunk_alley__agent (1 result), Cycling Peloton (14 results), svg_detective_noir__agent (1 result), Festival Oil Lamps (10 results), DNA Double Helix (10 results), Energy Mix Donut Chart (10 results), svg_dragon_sleeping_castle__agent (1 result), 3D Block Letters (10 results), Drum Kit (10 results), Basket of Painted Eggs (10 results), Fantasy World Map (10 results), svg_film_camera__agent (1 result), Coffee Decision Flowchart (10 results), svg_food_pyramid__agent (1 result), Food Pyramid (10 results), Fractal Tree (10 results), Straw in a Glass (10 results), Glitch Text (10 results), svg_grand_piano__agent (1 result), Grand Piano (9 results), svg_halloween_drippy_boo__agent (1 result), Drippy BOO Lettering (7 results), svg_hand_five_fruits__agent (1 result), Hand Holding Five Fruits (7 results), svg_hello_neon_sign__agent (1 result), Human Eye Diagram (7 results), svg_human_heart_anatomy__agent (1 result), Human Heart Anatomy (7 results), svg_human_skeleton_front__agent (1 result), Human Skeleton (7 results), svg_islamic_star_tiling__agent (1 result), Geometric Star Tiling (7 results), Isometric Coffee Shop (7 results), Isometric Cube Illusion (7 results), Isometric Farm (7 results), Isometric Harbour (7 results), Isometric Workspace (7 results), Isometric Space Station (7 results), svg_jack_o_lanterns__agent (1 result), Jack-o'-Lanterns (7 results), svg_japanese_pagoda__agent (1 result), Seigaiha Wave Pattern (7 results), svg_latte_art__agent (1 result), Temperature Line Chart (4 results), Bakery Badge (4 results), Bookshop Logo (4 results), Craft Drinks Label Emblem (4 results), svg_logo_coffee_roaster__agent (1 result), Coffee Roaster Logo (4 results), Gym Logo (4 results), Florist Monoline Logo (4 results), Outdoor Gear Emblem (4 results), Pet Clinic Logo (4 results), Space Program Patch (4 results), svg_logo_sushi_bar__agent (1 result), Sushi Bar Logo (4 results), Geometric Tech Logo (4 results), Lunar New Year Lanterns (4 results), Colourful Mandala (4 results), Memphis Pattern (2 results), Cat Seeing a Lion (2 results), Ornate Monogram (2 results), Phases of the Moon (2 results), svg_mountain_lake_dawn__agent (1 result), Hiking Trail Map (2 results), New Year Fireworks (2 results), Op Art Waves (2 results), Orchestra Seating Chart (2 results), Organisation Chart (2 results), Impossible Triangle (2 results), Periodic Table Excerpt (2 results), Budget Pie Chart (2 results), Plant Cell Diagram (2 results), Reflection in a Puddle (2 results), Recycling Process (2 results), svg_retro_crt_computer__agent (1 result), svg_rice_terraces__agent (1 result), Rock Band on Stage (2 results), Rock Climber (2 results), Saxophone (2 results), svg_sewing_machine__agent (1 result), Sheet Music Excerpt (2 results), Downhill Skier (2 results), Overhead Kick (2 results), Solar System (2 results), Sports Team Crest (2 results), Cherry Blossom Picnic (2 results), Abstract Stained Glass (2 results), Stay Wild Circular Badge (2 results), Metro Map (2 results), Summer Beach Day (2 results), Sunburst Rays (2 results), Surfer in a Barrel (2 results), svg_sushi_platter__agent (1 result), svg_suspension_bridge_fog__agent (1 result), Butterfly Stroke Swimmer (2 results), svg_swiss_army_knife__agent (1 result), Modular Synthesizer (2 results), Tennis Serve (2 results), Thank You Script (2 results), Harvest Dinner Table (2 results), Stacked Animals (2 results), History of Flight Timeline (2 results), svg_treasure_map__agent (1 result), Pirate Treasure Map (2 results), Valentine Letter (2 results), Three-Circle Venn Diagram (2 results), Vintage Travel Poster Type (2 results), svg_vinyl_turntable__agent (1 result), Record Turntable (2 results), Violin and Bow (2 results), Volcano Cross-Section (2 results), svg_voronoi_cells__agent (1 result), Voronoi Pattern (2 results), svg_water_cycle_diagram__agent (1 result), Water Cycle Diagram (2 results), svg_weather_widget__agent (1 result), Snowy Holiday Cabin (2 results), Cloud Lettering (2 results), svg_soccer_bicycle_kick__agent (1 result)
By model
Every recorded attempt counts, failures included (runs lost to infrastructure are not results and are left out). Medians use finished runs only; n is how many of them recorded the figure. Prompts, harnesses and retries differ between models, so these describe the runs rather than rank the models.
Show the table (23 rows)
| GPT-6 Sol | 127 | 127 | 100% | $0.34 n=15 | 1m 51s n=127 |
|---|---|---|---|---|---|
| GPT-6 Luna | 113 | 113 | 100% | not recorded: no finished run with a recorded cost | 1m 02s n=113 |
| Space Bunny Alpha | 21 | 21 | 100% | $0 n=20 | 5m 43s n=21 |
| Grok 4.7 | 10 | 10 | 100% | $6.29 n=10 | 36m 49s n=10 |
| Claude Sonnet 5 | 79 | 78 | 99% | not recorded: no finished run with a recorded cost | 2m 18s n=78 |
| GPT-6 Astra | 138 | 136 | 99% | $0.62 n=136 | 3m 47s n=136 |
| DeepSeek V4 Pro | 133 | 131 | 98% | $0.052 n=1 | 2m 20s n=131 |
| Claude Fable 5.1 | 110 | 107 | 97% | $0.80 n=107 | 3m 10s n=107 |
| Kimi K3 | 64 | 62 | 97% | $0.92 n=24 | 13m 23s n=62 |
| Muse Spark 1.3 | 24 | 23 | 96% | $0 n=18 | 5m 20s n=23 |
| gpt-oss-120b | 240 | 225 | 94% | not recorded: no finished run with a recorded cost | 8.0 s n=225 |
| LongCat 2.5 Preview | 46 | 42 | 91% | $0 n=42 | 9m 18s n=42 |
| DeepSeek V4.1 Flash | 173 | 157 | 91% | $0.083 n=24 | 1m 02s n=157 |
| Claude Opus 5.5 | 156 | 131 | 84% | $0.61 n=131 | 5m 01s n=131 |
| Qwen3.8 Max | 55 | 46 | 84% | not recorded: no finished run with a recorded cost | 3m 48s n=46 |
| MiniMax M3 | 49 | 36 | 73% | not recorded: no finished run with a recorded cost | 2m 29s n=36 |
| Claude Sonnet 5.5 | 17 | 9 | 53% | $6.66 n=9 | 32m 10s n=9 |
| Gemini 3.8 Flash | 16 | 8 | 50% | $0 n=3 | 9m 44s n=8 |
| Qwen3.8 27B | 238 | 112 | 47% | not recorded: no finished run with a recorded cost | 24 s n=112 |
| GLM-5.3 | 86 | 37 | 43% | $0.96 n=24 | 16m 08s n=37 |
| MiMo V2.6 Flash | 16 | 2 | 12% | $0 n=2 | 49m 26s n=2 |
| Gemini 3.6 Flash | 9 | 1 | 11% | $0 n=1 | 3m 19s n=1 |
| Gemini 3.7 Flash | 2 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
By harness
Show the table (14 rows)
| Codex CLI | 25 | 25 | 100% | $0.74 n=25 | 5m 22s n=25 |
|---|---|---|---|---|---|
| Kimi Code CLI | 12 | 12 | 100% | $1.13 n=12 | 19m 49s n=12 |
| Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills) | 6 | 6 | 100% | $2.56 n=6 | 14m 42s n=6 |
| Pi 0.87.1 | 5 | 5 | 100% | $4.89 n=5 | 29m 04s n=5 |
| Grok Build | 4 | 4 | 100% | $4.50 n=4 | 38m 25s n=4 |
| Kilo Code CLI 7.7.9 | 4 | 4 | 100% | $6.27 n=4 | 30m 28s n=4 |
| OpenCode | 162 | 149 | 92% | $0.18 n=148 | 13m 27s n=149 |
| Raw API call | 1542 | 1314 | 85% | $0.55 n=280 | 1m 24s n=1314 |
| dsh | 13 | 11 | 85% | $0.084 n=11 | 10m 40s n=11 |
| Oh My Pi (omp) 18.3.1 | 6 | 5 | 83% | $6.65 n=5 | 32m 54s n=5 |
| Claude Code | 64 | 45 | 70% | $6.28 n=45 | 28m 32s n=45 |
| Cline CLI | 65 | 33 | 51% | $0 n=21 | 5m 34s n=33 |
| Hermes Agent 0.21.5 | 8 | 1 | 12% | $23.83 n=1 | 2h 30m n=1 |
| Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5 | 6 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
Three.js
369 results from 85 prompts, 277 finished. Browse them
AI judge, per prompt
AI judge: the test runner (model claude-opus-5-5), scoring by visual review of the 1600 px render against the pre-registered rubric. Ranks order the recorded assessments within one prompt; equal totals share a rank and are provisional. Small differences have not been tested for repeatability. Only each cell's latest attempt is ranked; earlier attempts keep their pages. How results are scored
No prompt in this category has a published rubric yet, so nothing is ranked.
Unscored prompts (no published rubric yet, so no ranks): Fjord scene (Three.js) (21 results), Animated short (HTML/JS) (8 results), AI data centre, inside view (2 results), Santa Monica particle beach (2 results), folding_road_music (1 result), folding_road_threejs (13 results), gpu_fluid_sim (9 results), pixel_lofi_loop (4 results), threejs_dragon_ship (5 results), threejs_eiffel (8 results), threejs_ant_vs_gardener (8 results), Garden Automation Tutorial (2 results), threejs_jellyfish (8 results), threejs_momijigari (8 results), Night Train (9 results), Paris Tower (6 results), threejs_space_bunny (6 results), threejs_tidewater (7 results), x16_engine (11 results), Blizzard Cabin (1 result), cine_dolly_zoom (4 results), creature_robot_blueprint (6 results), html_fox_forest_run (4 results), humanoid_motion_cycle (6 results), threejs_procedural_armor_suit (6 results), cine_dutch_angle (2 results), cine_noir_low_key (4 results), cine_orbit_360 (4 results), cine_silhouette_backlight (4 results), cine_tracking_shot (3 results), game_battleship (6 results), threejs_slime_softbody (5 results), threejs_artwork_style_pass (11 results), threejs_mechanical_bee (11 results), threejs_stillwater_pond (7 results), threelab_breed_generative_art (2 results), webgpu_deferred_sponza_playground (1 result), webgpu_fishing_game (1 result), webgpu_ocean_wave_physics (2 results), webgpu_sailing_storm (1 result), one_prompt_3d_kart_racer (2 results), webgpu_particle_life_audio (1 result), webgpu_snow_physics (1 result), threejs_domino_chain_physics (1 result), threejs_evolving_art_garden (2 results), threejs_ps1_horror_kitchen (2 results), webgpu_ink_marbling (1 result), shader_hero_background_studio (2 results), threejs_tiny_anime_world (2 results), threejs_torchlit_dungeon_dragon (1 result), webgpu_ascii_shader_scene (2 results), one_key_rhythm_runner (2 results), threejs_marble_run_machine (1 result), threejs_solar_sail_orrery (2 results), webgl_reaction_diffusion_coral (2 results), image_to_procedural_threejs_scene (2 results), js13k_style_shader_game (3 results), procedural_lego_builder (1 result), threejs_boids_dragon (1 result), threejs_pathtraced_product_shot (2 results), threejs_rocket_car_arena (2 results), volumetric_raymarched_sun (2 results), webgl_particle_life_sim (2 results), Victorian optical toys in 3D: zoetrope, phenakistiscope, praxinoscope (Three.js) (2 results), Vintage biplane aerobatics over a grass airfield (Three.js) (4 results), Retro arcade claw machine with pastel plush prizes (Three.js) (6 results), Four-stroke engine cutaway: crank, piston and cam-driven valves (Three.js) (8 results), Hot air balloon festival over rolling fields (Three.js) (9 results), Synthwave hover-bike racer on a neon grid highway (Three.js) (7 results), Outer-space pinball table with planet bumpers (Three.js) (5 results), Floating pirate island diorama with a cliff cross-section (Three.js) (7 results), Rube Goldberg breakfast machine on a kitchen countertop (Three.js) (1 result), cine_crane_rise (1 result), cine_low_angle_hero (1 result), cine_rack_focus (1 result), cine_whip_pan (1 result), threejs_gerstner_ocean_storm (1 result), threejs_gpgpu_galaxy_collision (1 result), threejs_infinite_terrain_flyover (1 result), threejs_jetski_ocean_physics (1 result), threejs_soap_bubble_shader_playground (1 result), threejs_toy_pool_water (1 result), threejs_mobile_touch_arcade_game (4 results), threejs_octopus_reef (1 result), threejs_impressionist_world (1 result)
By model
Every recorded attempt counts, failures included (runs lost to infrastructure are not results and are left out). Medians use finished runs only; n is how many of them recorded the figure. Prompts, harnesses and retries differ between models, so these describe the runs rather than rank the models.
Show the table (20 rows)
| GPT-6 Sol | 42 | 42 | 100% | $1.51 n=40 | 10m 12s n=42 |
|---|---|---|---|---|---|
| qwen3p8-max | 8 | 8 | 100% | not recorded: no finished run with a recorded cost | 7m 43s n=8 |
| Claude Haiku 5.5 | 5 | 5 | 100% | not recorded: no finished run with a recorded cost | 8m 07s n=5 |
| gpt-oss-120b | 4 | 4 | 100% | not recorded: no finished run with a recorded cost | 14 s n=4 |
| Gemini 3.6 Flash | 2 | 2 | 100% | $0 n=2 | 2m 41s n=2 |
| GPT-6 Luna | 2 | 2 | 100% | not recorded: no finished run with a recorded cost | 2m 48s n=2 |
| GPT-6 Astra | 41 | 39 | 95% | $5.05 n=39 | 25m 14s n=39 |
| DeepSeek V4 Pro | 11 | 10 | 91% | not recorded: no finished run with a recorded cost | 4m 42s n=10 |
| gpt-6.1-sol | 21 | 19 | 90% | not recorded: no finished run with a recorded cost | 6m 32s n=19 |
| minimax-m3 | 8 | 7 | 88% | not recorded: no finished run with a recorded cost | 3m 37s n=7 |
| DeepSeek V4.1 Flash | 20 | 15 | 75% | $0.15 n=14 | 31m 19s n=15 |
| Claude Sonnet 5.5 | 33 | 24 | 73% | $0.84 n=1 | 24m 21s n=24 |
| Claude Opus 5.5 | 98 | 67 | 68% | $4.01 n=67 | 31m 18s n=67 |
| Claude Fable 5.1 | 37 | 21 | 57% | $14.17 n=21 | 50m 56s n=21 |
| gemini-3.5-flash | 2 | 1 | 50% | not recorded: no finished run with a recorded cost | 2m 15s n=1 |
| Space Bunny Alpha | 2 | 1 | 50% | $0 n=1 | 4m 55s n=1 |
| GLM-5.3 | 13 | 5 | 38% | $2.89 n=5 | 58m 38s n=5 |
| Kimi K3 | 11 | 4 | 36% | $6.01 n=4 | 1h 40m n=4 |
| Qwen3.8 27B | 3 | 1 | 33% | not recorded: no finished run with a recorded cost | 19 s n=1 |
| LongCat 2.5 Preview | 6 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
By harness
| Cline CLI | 1 | 1 | 100% | $0 n=1 | 4m 55s n=1 |
|---|---|---|---|---|---|
| Codex CLI | 66 | 64 | 97% | $1.78 n=55 | 16m 18s n=64 |
| Raw API call | 99 | 79 | 80% | $0.71 n=29 | 4m 43s n=79 |
| Claude Code | 101 | 80 | 79% | $4.01 n=56 | 26m 08s n=80 |
| dsh | 7 | 5 | 71% | $0.19 n=5 | 45m 32s n=5 |
| Kimi Code CLI | 3 | 2 | 67% | $8.76 n=2 | 1h 55m n=2 |
| OpenCode | 82 | 45 | 55% | $6.88 n=45 | 52m 30s n=45 |
| Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills) | 6 | 1 | 17% | $4.78 n=1 | 49m 01s n=1 |
| Hermes Agent 0.21.5 | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
| Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5 | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
| Kilo Code CLI 7.7.9 | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
| Pi 0.87.1 | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
HTML animation
459 results from 106 prompts, 354 finished. Browse them
AI judge, per prompt
AI judge: the test runner (model claude-opus-5-5), scoring by visual review of the 1600 px render against the pre-registered rubric. Ranks order the recorded assessments within one prompt; equal totals share a rank and are provisional. Small differences have not been tested for repeatability. Only each cell's latest attempt is ranked; earlier attempts keep their pages. How results are scored
No prompt in this category has a published rubric yet, so nothing is ranked.
Unscored prompts (no published rubric yet, so no ranks): Animated short (HTML/JS) (11 results), Van Gogh cat-spotting game (2 results), 16-bit pixel wizard sprite loop (2 results), "Make a Tron game" (2 results), 001 Ink Bloom Garden (14 results), 002 Pendulum Wave (14 results), 003 Orbital Coffee Landing (14 results), 004 Brick Breaker Neon (14 results), 005 Focus Timer Orbit (13 results), 006 Rainfall Bars (11 results), 007 Kinetic Headline (11 results), 008 Pocket Theremin (11 results), 009 Forest Fire Automaton (11 results), 010 Toggle Switch Gallery (11 results), 011 Pure CSS Sunset (11 results), 012 Magnetic Filings (11 results), 013 Soft Body Blobs (11 results), 014 Snake Reimagined (11 results), 015 Color Palette Studio (11 results), 016 Heartbeat Line (11 results), 017 Typewriter Poem (11 results), 018 Step Sequencer Grid (11 results), 019 Boids Lagoon (8 results), 020 Pricing Table Trio (6 results), 021 Rotating CSS Cube (5 results), 022 Mandala Painter (5 results), 023 Mortgage Calculator (4 results), 024 Tower Stack (3 results), 025 Gravity Well (3 results), 026 Aurora Ribbons (2 results), 027 Momentum Cradle (2 results), 028 Night Market Zine (2 results), 029 Asteroid Drift (2 results), 030 Unit Converter Deck (2 results), 031 Population Bubbles (2 results), 032 Liquid Letters (2 results), 033 Wind Chimes (2 results), 034 Sandpile Falling (2 results), 035 Modal Dialog Set (2 results), 036 CSS Jellyfish Tank (2 results), 037 Voronoi Stained Glass (2 results), 038 Cloth Curtain (2 results), 039 Memory Match Cards (2 results), 040 Markdown Notepad (2 results), 041 Solar Radial Clock (2 results), 042 Split Flap Board (2 results), 043 Drone Pad Synth (2 results), 044 Traffic Intersection (2 results), 045 Form Validation Demo (2 results), 046 Pure CSS Loaders (2 results), 047 Flow Field Threads (2 results), 048 Marble Run Ramps (2 results), 049 Maze Runner (2 results), 050 Budget Pie Planner (2 results), 051 Tidal Moiré (2 results), 052 Catapult Castle (2 results), 053 Portfolio Grid (2 results), 054 Flappy Paper Plane (2 results), 055 Password Forge (2 results), 056 Commute Heatmap (2 results), 057 Variable Weight Wave (2 results), 058 Bubble Wrap Pop (2 results), 059 Predator Prey Meadow (2 results), 060 Segmented Nav Kit (2 results), 061 CSS Candle Flame (2 results), 062 Recursive Tree Wind (2 results), 063 Rope Bridge (2 results), 064 Minesweeper Classic (2 results), 065 Habit Streak Tracker (2 results), 066 Stock Candlesticks (2 results), 067 Glitch Title Card (2 results), 068 Piano Keys Mini (2 results), 069 Ant Colony Trails (2 results), 070 Card Hover Effects (2 results), 071 CSS Pixel Portrait (2 results), 072 Starfield Warp (2 results), 073 Water Ripple Pond (2 results), 074 Word Guess Five (2 results), 075 Tip Splitter (2 results), 076 Reaction Diffusion (2 results), 077 Double Pendulum Chaos (2 results), 078 Event Poster Page (2 results), 079 Tetromino Drop (2 results), 080 Regex Playground (2 results), 081 Election Hemicycle (2 results), 082 Circular Text Badge (2 results), 083 Sine Chord Pad (2 results), 084 Epidemic Grid (2 results), 085 Dropdown Menu Kit (2 results), 086 CSS Ocean Waves (2 results), 087 Particle Text Morph (2 results), 088 Spring Mass Chain (2 results), 089 Pixel Paint Paddles (2 results), 090 Countdown Builder (2 results), 091 Treemap Budget (2 results), 092 Scroll Story Words (2 results), 093 Arpeggio Machine (2 results), 094 Lava Lamp Metaballs (2 results), 095 Notification Centre (2 results), 096 CSS Solar System (2 results), 097 Wave Interference (1 result), Zoom to the Atom (4 results), Koi Pond Garden (12 results), coastal_shallows (1 result), Codrops-style scroll story: the life of a raindrop (5 results), firstprinciples_mascot_multimedia (1 result)
By model
Every recorded attempt counts, failures included (runs lost to infrastructure are not results and are left out). Medians use finished runs only; n is how many of them recorded the figure. Prompts, harnesses and retries differ between models, so these describe the runs rather than rank the models.
Show the table (17 rows)
| GPT-6 Sol | 26 | 26 | 100% | $0.61 n=6 | 2m 38s n=26 |
|---|---|---|---|---|---|
| GPT-6 Astra | 22 | 22 | 100% | $0.85 n=22 | 4m 51s n=22 |
| GPT-6 Luna | 19 | 19 | 100% | not recorded: no finished run with a recorded cost | 1m 53s n=19 |
| gpt-6.1-sol | 1 | 1 | 100% | not recorded: no finished run with a recorded cost | 2m 40s n=1 |
| Space Bunny Alpha | 1 | 1 | 100% | $0 n=1 | 9m 22s n=1 |
| DeepSeek V4 Pro | 27 | 26 | 96% | not recorded: no finished run with a recorded cost | 2m 20s n=26 |
| Kimi K3 | 26 | 25 | 96% | not recorded: no finished run with a recorded cost | 6m 03s n=25 |
| Claude Fable 5.1 | 23 | 22 | 96% | $0.82 n=22 | 2m 57s n=22 |
| Claude Sonnet 5 | 20 | 19 | 95% | not recorded: no finished run with a recorded cost | 2m 10s n=19 |
| gpt-oss-120b | 110 | 95 | 86% | not recorded: no finished run with a recorded cost | 9.8 s n=95 |
| DeepSeek V4.1 Flash | 26 | 22 | 85% | not recorded: no finished run with a recorded cost | 1m 41s n=22 |
| Claude Opus 5.5 | 33 | 27 | 82% | $0.51 n=27 | 3m 29s n=27 |
| minimax-m3 | 7 | 5 | 71% | not recorded: no finished run with a recorded cost | 2m 29s n=5 |
| Qwen3.8 Max | 8 | 5 | 62% | not recorded: no finished run with a recorded cost | 3m 29s n=5 |
| Qwen3.8 27B | 99 | 37 | 37% | not recorded: no finished run with a recorded cost | 31 s n=37 |
| Claude Sonnet 5.5 | 4 | 1 | 25% | not recorded: no finished run with a recorded cost | 22m 48s n=1 |
| GLM-5.3 | 7 | 1 | 14% | not recorded: no finished run with a recorded cost | 18m 33s n=1 |
By harness
| Codex CLI | 7 | 7 | 100% | $1.15 n=7 | 8m 54s n=7 |
|---|---|---|---|---|---|
| Cline CLI | 1 | 1 | 100% | $0 n=1 | 9m 22s n=1 |
| Raw API call | 437 | 337 | 77% | $0.65 n=62 | 1m 36s n=337 |
| Claude Code | 11 | 8 | 73% | $4.50 n=7 | 19m 50s n=8 |
| OpenCode | 2 | 1 | 50% | $8.58 n=1 | 25m 53s n=1 |
| Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills) | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
Blender
28 results from 11 prompts, 24 finished. Browse them
AI judge, per prompt
AI judge: the test runner (model claude-opus-5-5), scoring by visual review of the 1600 px render against the pre-registered rubric. Ranks order the recorded assessments within one prompt; equal totals share a rank and are provisional. Small differences have not been tested for repeatability. Only each cell's latest attempt is ranked; earlier attempts keep their pages. How results are scored
No prompt in this category has a published rubric yet, so nothing is ranked.
Unscored prompts (no published rubric yet, so no ranks): blender_donut (3 results), blender_lowpoly_dragon (5 results), blender_chess_set_midgame (1 result), blender_cosy_isometric_study (1 result), blender_lighthouse_diorama (1 result), blender_sci_fi_helmet (1 result), blender_vintage_rangefinder_camera (1 result), blender_steam_locomotive (1 result), quadruped_rig_walk (1 result), blender_humanoid_walk_cycle (6 results), blender_spongebob (6 results)
By model
Every recorded attempt counts, failures included (runs lost to infrastructure are not results and are left out). Medians use finished runs only; n is how many of them recorded the figure. Prompts, harnesses and retries differ between models, so these describe the runs rather than rank the models.
| GPT-6 Astra | 11 | 11 | 100% | $9.48 n=11 | 37m 23s n=11 |
|---|---|---|---|---|---|
| gpt-6.1-sol | 2 | 2 | 100% | not recorded: no finished run with a recorded cost | 54m 14s n=2 |
| GPT-6 Sol | 1 | 1 | 100% | $1.64 n=1 | 17m 55s n=1 |
| Kimi K3 | 1 | 1 | 100% | $2.31 n=1 | 1h 21m n=1 |
| DeepSeek V4.1 Flash | 5 | 4 | 80% | $0.15 n=4 | 1h 17m n=4 |
| Claude Opus 5.5 | 4 | 3 | 75% | $8.85 n=3 | 1h 22m n=3 |
| Claude Sonnet 5.5 | 2 | 1 | 50% | not recorded: no finished run with a recorded cost | 44m 56s n=1 |
| GLM-5.3 | 2 | 1 | 50% | $1.70 n=1 | 1h 06m n=1 |
By harness
| Codex CLI | 13 | 13 | 100% | $9.48 n=11 | 37m 20s n=13 |
|---|---|---|---|---|---|
| dsh | 1 | 1 | 100% | $0.17 n=1 | 1h 07m n=1 |
| OpenCode | 8 | 6 | 75% | $0.93 n=6 | 1h 17m n=6 |
| Claude Code | 6 | 4 | 67% | $8.85 n=3 | 1h 03m n=4 |
Video
75 results from 29 prompts, 58 finished. Browse them
AI judge, per prompt
AI judge: the test runner (model claude-opus-5-5), scoring by visual review of the 1600 px render against the pre-registered rubric. Ranks order the recorded assessments within one prompt; equal totals share a rank and are provisional. Small differences have not been tested for repeatability. Only each cell's latest attempt is ranked; earlier attempts keep their pages. How results are scored
No prompt in this category has a published rubric yet, so nothing is ranked.
Unscored prompts (no published rubric yet, so no ranks): hyperframes_artemis_ii (14 results), remotion_sales_chart (2 results), bacterium_to_quantum_zoom (4 results), hyperframes_glass_launch_ui (4 results), remotion_ai_dungeon (3 results), remotion_how_internet_works (3 results), video_printing_press_collage_hyperframes (6 results), video_printing_press_collage_remotion (5 results), hyperframes_brutalist_event_promo (2 results), hyperframes_css_3d_cube_gallery (1 result), hyperframes_glass_dashboard_launch (2 results), hyperframes_ink_brush_dragon (2 results), hyperframes_isometric_server_room (2 results), hyperframes_lighthouse_storm (2 results), hyperframes_product_reveal_watch (1 result), hyperframes_terminal_hacker_boot (2 results), hyperframes_tokyo_neon_title (1 result), remotion_chess_famous_game (1 result), remotion_chip_fabrication_journey (1 result), remotion_energy_mix_sankey (2 results), remotion_f1_lap_telemetry (2 results), remotion_galaxy_collision (2 results), remotion_github_contrib_wrapped (1 result), remotion_music_visualizer_synthwave (2 results), remotion_orbital_launch_sequence (1 result), remotion_satellite_constellation (2 results), remotion_sorting_algorithms_race (1 result), video_style_morph_character_hyperframes (1 result), video_style_morph_character_remotion (1 result)
By model
Every recorded attempt counts, failures included (runs lost to infrastructure are not results and are left out). Medians use finished runs only; n is how many of them recorded the figure. Prompts, harnesses and retries differ between models, so these describe the runs rather than rank the models.
| GPT-6 Sol | 19 | 19 | 100% | $2.10 n=19 | 14m 28s n=19 |
|---|---|---|---|---|---|
| Claude Fable 5.1 | 3 | 3 | 100% | $40.74 n=3 | 1h 39m n=3 |
| GPT-6 Astra | 3 | 3 | 100% | $16.60 n=3 | 55m 41s n=3 |
| gpt-6.1-sol | 2 | 2 | 100% | not recorded: no finished run with a recorded cost | 35m 34s n=2 |
| Kimi K3 | 1 | 1 | 100% | $2.09 n=1 | 51m 54s n=1 |
| Claude Opus 5.5 | 28 | 23 | 82% | $5.19 n=23 | 33m 38s n=23 |
| DeepSeek V4.1 Flash | 5 | 3 | 60% | $0.22 n=3 | 1h 44m n=3 |
| GLM-5.3 | 2 | 1 | 50% | $4.76 n=1 | 2h 47m n=1 |
| Claude Sonnet 5.5 | 12 | 3 | 25% | not recorded: no finished run with a recorded cost | 37m 40s n=3 |
By harness
| Codex CLI | 23 | 23 | 100% | $2.63 n=21 | 15m 39s n=23 |
|---|---|---|---|---|---|
| Pi 0.87.1 | 2 | 2 | 100% | $12.18 n=2 | 1h 10m n=2 |
| dsh | 1 | 1 | 100% | $0.22 n=1 | 1h 33m n=1 |
| Kilo Code CLI 7.7.9 | 1 | 1 | 100% | $33.35 n=1 | 2h 18m n=1 |
| Oh My Pi (omp) 18.3.1 | 1 | 1 | 100% | $32.29 n=1 | 2h 24m n=1 |
| Claude Code | 28 | 21 | 75% | $5.14 n=18 | 33m 38s n=21 |
| Pi 0.87.1 + skills kit (ponytail, ECC, vetted skills) | 3 | 2 | 67% | $12.27 n=2 | 1h 19m n=2 |
| OpenCode | 14 | 7 | 50% | $4.76 n=7 | 2h 01m n=7 |
| Hermes Agent 0.21.5 | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
| Hermes Agent 0.21.5 + Oh-My-Hermes 2.0.5 | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |