Opus 5 reportedly improvised a pseudo-vision stack on a blind FreeCAD eval
teortaxesTex · x · 2026-07-25
Opus 5 reportedly built its own pseudo-vision stack on a blind engineering task
A quoted eval says Claude Opus 5 was tested on a difficult benchmark task: reconstructing a machine part as a 3D FreeCAD model from an engineering drawing, while only seeing terminal text in a text-only mini-swe-agent harness.
- The model first tried standard OCR with Pillow and Tesseract.
- Thin numerals on the drawing were too unreliable, so it improvised a custom vision stack.
- The thread’s point is that the impressive part is not just the success, but that the model can assemble tools on the fly when deprived of direct vision.
A reply pushes back on framing this as Anthropic-specific, arguing that weaker blind models can also build ad hoc "pseudo-vision" from available tools; the real standout is still solving the multidisciplinary task.
Related event: Weak Vision Models Exposed for Faking Visual Abilities with Tools(2 posts)→
More from coding & agent
- Codex tip: use Sol with Astra and Luna sub-agents to save usage — pvncher · 2026-09-11
- agents-best-practices: a provider-neutral Agent Skill for designing and auditing agentic harnesses — tom_doerr · 2026-09-11
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- First-ever Three.js Conference lands in Paris, with a panel on AI-shortened design workflows — OdinLovis · 2026-09-11
- Data engineering, not agent frameworks, is the real bottleneck for enterprise AI agents — dhruv2038 · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11