Claude Opus 4.7 Operates Nonstop for 19 Days on MirrorCode Task at $2,600 Cost

Claude Opus 4.7 operated continuously for 19 days on a single MirrorCode task, costing $2,600. Developed by Epoch AI, the model achieved a 56% solve rate while rebuilding a toolkit of 16,000 lines in 14 hours. Despite this performance, it was noted that ‘every model tested still fails on the most complex tasks,’ highlighting ongoing challenges in AI capabilities. This follows previous reports on Anthropic’s models facing bans and economic implications in the AI sector, underscoring the competitive landscape in advanced AI development. Practitioners can analyze Claude Opus 4.7’s performance metrics to inform their own model training and benchmarking strategies. For further details, see The Decoder.