Donatello AI Benchmark / Protocol & evidence
Versions, corrections and what changed.
A public record of protocol and data releases. Future results will preserve their original input version, model cohort and execution dates.
Protocol v1 and its downloads remain available. The separately registered October 6 pilot adds 24 API attempts, technical results and costs. No public votes, independently reviewed quality scores or quality winners are published.
2026-10-06 · 2026-09-11-v1-restoration
Historical Creator Value download restored and decoupled from live prices
The October 6 price update had unintentionally recalculated the historical v1 Creator Value JSON and CSV with today's fixed 3-credit image tariff. Restored the original v1 calculation (1–8 image credits) and its original file checksums. The downloadable archive now uses a separate frozen calculation; today's platform comparisons continue to use current tariffs. This restores the original record, not a new measured result. The current interactive explorer uses the archived platform snapshot with the October 6 fixed image-credit calculation, as labelled on its page.
Inspect this release's versioned files →2026-10-06 · 2026-10-06-pilot-v1
Separate four-prompt API pilot — technical measurements only
Published all 24 scheduled attempts for FLUX.2 [pro] and Ideogram 4.5 medium via Runware, including costs, response times, input hashes and original outputs. The manifest was frozen locally before dispatch, not independently preregistered. This pilot measures dispatch-to-JSON response time, excluding later downloads; it does not use the original protocol's end-to-end latency definition. No independent usability review, quality score or platform winner is claimed. Original v1 prompt, reference and Creator Value downloads remain unchanged.
Inspect this release's versioned files →2026-09-11 · 2026-09-11-v1
Protocol and documentation release — no measured leaderboard
Published 30 image prompts, eight reference cases (five awaiting cleared references), the measurement contract and a documented Creator Value explorer. Zero comparative generation runs or public votes. No quality winner, completion rate or usable-output cost is claimed.
Inspect this release's versioned files →