M3 · Claude Sonnet 4.6 SWE-bench Verified — published 2026-05-10, frozen at market creation

Resolution spec: Claude Sonnet 4.6 will score > 88 on SWE-bench Verified by 2026-06-15

Market parameters

Resolution rule

Outcome resolves to YES if, at the snapshot timestamp 2026-06-15T12:00:00Z, the SWE-bench Verified public leaderboard contains at least one submission attributed to "Claude Sonnet 4.6" (or equivalent: "claude-sonnet-4-6", "Claude 4.6 Sonnet") with a verified score strictly greater than 88.0. Otherwise resolves to NO.

Data source (authoritative)

• URL: https://www.swebench.com/verified.html (SWE-bench Verified leaderboard)
• Snapshot: archived via archive.org Wayback Machine at 2026-06-15T12:00Z
• Backup snapshot: https://github.com/swe-bench/SWE-bench/tree/main/SWE-bench-Verified/results for the same date
• Score field: the "% Resolved" column (or its successor on the leaderboard)
• Attribution match: case-insensitive contains "claude" AND "4.6" AND "sonnet" in the submission name

Edge cases (frozen)

• If SWE-bench Verified is deprecated or the leaderboard URL is dead, outcome = Invalid
• If multiple submissions match: the highest score is used
• If the snapshot exactly equals 88.0 (not greater than), outcome = NO
• If Anthropic releases "Claude Sonnet 4.7" before resolvesAt with a score > 88 but no 4.6 submission ≥ 88, outcome = NO
• If oracle fails to broadcast resolve within 24h of resolvesAt, traders MAY dispute via on-chain Invalid flag

QED receipt contents

On resolve, a QED receipt PDF will be sealed with:

• Wayback Machine snapshot URL and HTML hash
• The matching row(s) from the leaderboard
• The numeric score used in the determination
• D-KaP hybrid signature over the payload
• The Base mainnet resolution tx hash

trade M3 →