💡 How to use these prompts:
Replace everything in [BRACKETS] with your specific details before sending.
Click Copy to copy any prompt to your clipboard instantly.
Read this model announcement. For every benchmark figure, state the exact score, the benchmark name and version, who ran the evaluation, what harness or configuration was used, and the reasoning or compute setting. Put each in a table. Where the page does not state one of those, write not stated rather than inferring it.
For each benchmark figure in this announcement, state whether the comparison models were run under the same harness, the same compute budget and the same tool access. List every figure where they were not. Those are the numbers that cannot be placed on a leaderboard.
List the competing models that a reader would expect to see in this comparison and that are absent. For each, say whether the page gives a reason. Then say which absence most changes how the results read.
From this announcement, work out or locate what it cost to produce each headline figure: compute budget, number of attempts allowed, tool calls permitted, and any best-of-n or ensemble configuration. State which figures depended on a setting a normal user could not afford or access.
Restate this announcement in plain language, keeping only claims the page itself supports under conditions comparable to other models. Mark every figure you dropped and say why. Keep it under 200 words.