VQV Signal
According to a post on X by @SciFi, a new benchmark called TreasuryQA Bench uses 70 questions over 1,435 scanned U.S. Treasury publications kept in a Dropbox twin. Agents must find files through Dropbox's MCP server and read the scans with OCR. Of ten agents tested, Claude Opus 5.5 in Claude Code solved 64.8% of attempts, while GPT-5.6 Terra in Codex solved 16.7%, and every agent performed worse on blurrier scans.
According to a post on X by @SciFi, a new benchmark called TreasuryQA Bench uses 70 questions over 1,435 scanned U.S. Treasury publications kept in a Dropbox twin. Agents must find files through Dropbox's MCP server and read the scans with OCR. Of ten agents tested, Claude Opus 5.5 in Claude Code solved 64.8% of attempts, while GPT-5.6 Terra in Codex solved 16.7%, and every agent performed worse on blurrier scans.
AI-assisted summary based on the listed source.
The new benchmark has 70 questions over 1,435 scanned U.S. Treasury publications kept in a Dropbox twin. Agents must find files through Dropbox's MCP server and read scans with OCR. Of ten agents tested, Claude Opus 5.5 in Claude Code solved 64.8% of attempts and GPT-5.6 Terra in Codex solved 16.7%, and every...
The reported results suggest that tasks like pulling answers from scanned financial records remain error-prone, even for the strongest agent tested. Teams building agents that work with file stores or OCR pipelines may want to note the drop in accuracy on lower-quality scans. Watch for the full benchmark details to see how the results were measured.
VQV organizes public signals from inspectable sources. It does not independently verify the underlying report.
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
VQV surfaced this signal because it is recent, relevant to AI Coding Tools, connected to Scobleizer News Pulse (X).
No login, cookies, social SDKs, or automatic posting.
