Compute each task cost from reported per-request usage, including cached input. Show per-task, total and average API-equivalent estimates and link official pricing. Preserve original run data and actual-billed unknown values.
Match ChemLab results to the Booster T2 viewer layout
Show the four model-driven tasks with videos, instructions, token usage and session/tool-call viewers. Remove scripted acceptance and developer panels from the public page. Preserve the existing evaluation data and recordings.
Show real GPT-6 Astra runs with synchronized instrument video and model traces
