DeepSeek's new open model does a coding task for about 20 cents

In DeepSeek's own chart, V4.1-Flash-1017 (released Oct 17) scores 77.0% on DeepSWE v1.1 at under $0.20 per task, vs 74.2% for the old Flash. These are DeepSeek's numbers, with no independent results yet. Open weights, MIT license, 763B parameters, 1M-token context.