News / Open source · Oct 18, 2026
DeepSeek updates V4.1 Flash with agentic gains, still MIT-licensed
DeepSeek's V4.1-Flash-1017 is a minor update with self-reported jumps on Terminal-Bench and Humanity's Last Exam, and shorter outputs.
DeepSeek published V4.1-Flash-1017 on October 17 and describes it as a minor update to V4.1-Flash. According to the model card as summarized by Aiholics, it improves reasoning, coding, agentic and visual tasks while shortening task time and output length.
The model has about 763 billion parameters and a 1M-token context window, and the MIT license covers the weights.
DeepSeek's self-reported scores include Terminal-Bench 4.0 at 42.3, up from 31.2, DeepSWE v1.1 at 77.0, and Humanity's Last Exam without tools at 42.6, up from 36.8. On several other tests the card shows Claude Opus 5 ahead. Aiholics' price tracker lists it at $0.15 per million input tokens and $0.60 per million output.
Why it matters for builders
A big jump on a terminal benchmark at very low per-token prices makes this a candidate for cheap background agents. The scores are self-reported, so run your own evals before switching.
Sources
Get stories like this every Monday
One tested AI workflow with the prompts, plus 3 news items that matter if you build. Free.
More AI news
- Mistral's Voxtral 3: transcription from $0.0035 a minute and a voice agent SDK built on LiveKit
- DeepSeek open-sources DeepInfer, the inference engine behind V4.1 Flash, under Apache 2.0
- Vercel's durable-harness lets Claude Code, Codex, OpenCode and Pi sessions survive a crash mid-turn
- Claude API adds user profiles, so a safety action can hit one end customer instead of your whole org