Launching today
DeepSeek V4.1 Flash 1017
DeepSeek’s faster, cheaper Flash model, plus a search API
2 followers
DeepSeek’s faster, cheaper Flash model, plus a search API
2 followers
DeepSeek-V4.1-Flash-1017 is a new post-training update to DeepSeek’s Flash model, served as deepseek-flash on the website, app and API. DeepSeek says it scores higher, finishes tasks faster and writes shorter outputs than the September version. It has a 1M-token context and is tuned for coding agents like DeepSeek Harness. It also adds a Search and Fetch API. Peak price is $0.30 in and $1.20 out per 1M tokens.
Free Options
Launch tags:Artificial Intelligence
Launch Team
Customer.ioAutomate Messaging Everywhere — Startups Get 12 Months Free
Promoted
DeepSeek-V4.1-Flash-1017 is a new post-training update to DeepSeek’s V4.1-Flash model. Per DeepSeek, it scores higher, finishes tasks faster and writes shorter outputs than the September version, which it calls a “Pareto improvement in performance and cost”.
If you haven’t used Flash before, it’s DeepSeek’s fast, low-cost model. It takes text and image input, has a 1M-token context with up to 384K tokens of output, and is served as deepseek-flash on the website, app and API.
What changed in this update:
DeepSWE: 74.2 → 77.0
Terminal-Bench 4.0: 31.2 → 42.3
Automation-Bench: 54.8 → 60.5
Apex Bench: 64.6 → 72.1
ProgramBench: 20.3 → 29.1
HLE (no tools): 36.8 → 42.6
Better on all 11 benchmarks DeepSeek reported, by 2.0 to 11.1 points
It’s also tuned for agent work. DeepSeek trained it with reinforcement learning with self-feedback for end-to-end projects, from game creation to 3D modeling, and optimized it for DeepSeek Harness (developer preview), Tencent CodeBuddy and WorkBuddy.
There’s a new Search and Fetch API too. It costs $0.75 per 1,000 successful requests for each endpoint, and calls through DeepSeek Harness, Claude Code or Codex are free until November 1, 2026.
The price stays low, at $0.30 input and $1.20 output per 1M tokens at peak, and half that off-peak. The calling method and the deepseek-flash name haven’t changed, so existing apps get the update without code changes. The old deepseek-v4-flash names are retired and now route to this model.
Good fit if you’re building coding agents or automation and want a cheap model with a long context.