DeepSeek Flash-1017: Benchmark Cost per Solved Task

On October 17 DeepSeek pushed DeepSeek-V4.1-Flash-1017 to Hugging Face under an MIT license, and within days it was the top AI story on Hacker News with roughly 950 points. The model card calls it a minor update. The numbers say otherwise: Terminal-Bench 4.0 went from 31.2 to 42.3, DeepSWE v1.1 from 74.2 to 77.0, and the release notes claim it gets there with shorter outputs and faster runs.

The part worth your attention is cost. The Decoder's reading of DeepSeek's DeepSWE chart puts the new model near $0.20 per task at its top setting and about $0.10 at a low setting, versus around $0.50 for the previous Flash. Those are vendor-side figures on DeepSeek's own harness, and independent benchmarks were not out when the coverage ran. So the useful move is not to repeat the claim. It is to measure the same thing on your own tasks.