GPT-6 Astra · Infinite cathedral
The episode contrasts the families' look and gives Astra about 26 minutes, Sol about 36 minutes, and Opus about 1 hour 40 minutes, with the Claude runs near two hours. Astra's display is 26m 21.4s. Sol's invocation wall is 36m 40s. Opus's generation wall is 1h 42m 31s. Sonnet's elapsed time is 1h 51m 33s. Pending and partial ledger states stay as recorded. Shader capture Unaided model Open this round
Measured harness ledgerPublic result
GPT-6 Astra

Infinite cathedral corridor shader — GPT-6 Astra Max

Generate an infinite stained-glass cathedral corridor shader with convincing depth, architectural repetition, light, and motion.

Max reasoningHeadline result
Workflow cost
$10.78
Wall-clock
26m 21.4s wall-clock
Processed tokens
7.64M processed
Record state
partial_root_task_metrics_with_model_supplied_shadertoy_runtime_evidence
Public summary

GPT-6 Astra Max partial_root_task_metrics_with_model_supplied_shadertoy_runtime_evidence ledger: 26m 21.4s wall-clock, 7.64M processed, and $10.78 Verified Standard API-equivalent token estimate; not a subscription invoice..

Run identity and stack
  • Result ID: infinite-cathedral-shadertoy-gpt-6-astra-max
  • Technical model: gpt-6-astra
  • Provider: OpenAI
  • Client: Codex
  • Stack: OpenAI
  • Stack: Codex
  • Stack: Technical model/configuration: gpt-6-astra
  • Stack: ShaderToy fragment shader
  • Stack: Harness v1 infinite cathedral prompt
Primary artifact integrity
  • Kind: shadertoy-image-pass-fragment-shader
  • Path: artifacts/infinite-cathedral-shadertoy-gpt-6-astra-max/source/infinite-cathedral.glsl
  • SHA-256: 47039df4e6dee5cb6f7fac968396b8e9e8d60be838d0baf05dfbfaa6af0bdce9
Validation evidence
  • Result: PARTIAL_MODEL_SUPPLIED_SHADERTOY_RUNTIME_EVIDENCE
Recorded caveats
  • The bounded root task is not strict one-shot and its derived timing includes root-task work outside a pure model-only generation interval.
  • Output includes hidden reasoning, visible response content, code, and tool-call JSON.
  • Cache reads are discounted, so total processed tokens overstate cost.
  • Cache creation is zero because the transcript schema does not expose a cache-write token field; this is not proof of no cache writes.
  • The local supplied runtime evidence and independent static compile are not a Shadertoy performance measurement.
  • No initial/optimized local FPS pair, RTX Pro 6000 1080p60 final capture, or blind evaluation is archived.
Visible evidence gaps
  • strict one-user-turn protocol
  • initial and optimized local FPS under identical uncapped conditions
  • RTX Pro 6000 1080p60 final capture with GPU and frame-rate metadata
  • blind evaluation record
Public result only

This result keeps its public summary and evidence, but it does not currently have a matching Builder test with prompts, projects, Harness workflows or skills.

RemakeBenchResearch console