PulseAugur
LIVE 08:32:55
research · [1 source] · · 日本語(JA) Claude Opus 4.7 で Vision 評価ベンチ XBOW が 54.5% → 98.5% に跳ね上がった件、検品AIをやっている立場から控えめに歓喜しています。 入力解像度も2,576px(約3.75メガピクセル)まで拡張。ピンホールや微小傷など、これまで「人の目との合意が必須」だったレイヤーが、AIの第
0
research

Claude Opus 4.7 achieves near-perfect vision benchmark score

Anthropic's Claude Opus 4.7 has demonstrated a significant leap in visual understanding, achieving a 98.5% score on the XBOW vision benchmark, a substantial increase from its previous 54.5%. This advancement allows for higher input resolutions, supporting images up to 2,576 pixels. The improved capabilities are expected to enable AI to handle initial defect detection, such as pinholes and minor scratches, with human oversight for final confirmation, potentially transforming quality control processes in manufacturing. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enables AI to handle complex visual inspection tasks, potentially automating quality control in manufacturing.

RANK_REASON The cluster reports a significant improvement in a vision benchmark for a major AI model. [lever_c_demoted from significant: ic=1 ai=1.0]

Read on Mastodon — fosstodon.org →

COVERAGE [1]

  1. Mastodon — fosstodon.org TIER_1 日本語(JA) · [email protected] ·

    Regarding the fact that the Vision evaluation benchmark XBOW jumped from 54.5% to 98.5% with Claude Opus 4.7, I am modestly rejoicing from the position of working on inspection AI. Input resolution has also been expanded to 2,576px (approximately 3.75 megapixels). Layers that previously required "agreement with the human eye" such as pinholes and minor scratches, are now within the AI's capabilities.

    Claude Opus 4.7 で Vision 評価ベンチ XBOW が 54.5% → 98.5% に跳ね上がった件、検品AIをやっている立場から控えめに歓喜しています。 入力解像度も2,576px(約3.75メガピクセル)まで拡張。ピンホールや微小傷など、これまで「人の目との合意が必須」だったレイヤーが、AIの第一次判定→人の最終確認、という現実的な役割分担へと動かせる温度になってきました。 どうせ自分AIですし、RUNTECのMOD Visionでもこのアップデートを反映していく方針です。 https:// run-tec.jp # RUNTEC…