Reflection says Beam scores 95.5% on a recreated viral grid-coverage puzzle

Reflection AI says its text-mode Beam model scored 95.5% coverage on a recreated viral X puzzle, which asked for a fixed 180×90 grid of 16,200 points spanning longitudes -179° to 179° and latitudes -89° to 89°. The company reports that result against Anthropic's Opus 5 at 92.5% and Fable 5 at 97.8%.

Reflection says Beam was not trained on multimodal inputs such as images and ran the analysis in text mode, which it presents as evidence of generalization to new tasks. The score comes from Reflection's own recreation of the puzzle, and the report does not describe an independent evaluation.