DeepSeek V1 Flash Just Got a Serious Upgrade

DeepSeek V4.1 Flash has been upgraded to 1017 on OpenCode Go and Free. No config changes needed. It’s still the same model family, but DeepSeek says this version reasons better, finishes tasks faster, and uses fewer output tokens. Official update (s'ouvre dans un nouvel onglet)

The gains are substantial on several coding tests: Terminal-Bench 4.0 rose from 31.2 to 42.3, and ProgramBench from 20.3 to 29.1. DeepSWE v1.1 improved from 74.2 to 77.0. Those are benchmark gains, not a guarantee every project will improve. Full results (s'ouvre dans un nouvel onglet)

It now reaches or beats some flagship-model scores on specific agent benchmarks, making “close to Opus-level” a fair test hypothesis—not a proven overall equivalence to Claude Opus 5.5. Test it on your own repo before switching important work over. DeepSeek says API pricing and calling method remain unchanged; OpenCode users can try it through existing Go or Free access.
We talked about this earlier.

PoorAI·PublicationDeepSeek V4.1 Flash Free on OpenCodeOpenCode Console lists DeepSeek V4.1 Flash Free at $0 per million input and output tokens. It is available alongside paid DeepSeek models, so check the selected model and pricing before running a large coding session. OpenCode Console models and pricing The model is also prominent in OpenCode’sLire la publication →

Worth retesting now: same setup, stronger reported coding results, no config change.