# DeepSeek V4.1-Flash-1017, Explained

> DeepSeek V4.1-Flash-1017 (released October 17, 2026) is a tuned checkpoint of V4.1 Flash, served under the same deepseek-flash API name at the same price: $0.15/$0.60 per 1M tokens. Includes an API cost calculator, a benchmark table, the new $0.75/1K Search & Fetch API rates, and how to switch.

DeepSeek V4.1-Flash-1017 went live on October 17, 2026. It is an updated post-training checkpoint of the V4.1 Flash model launched September 10, 2026. DeepSeek calls it a Pareto improvement in performance and cost: stronger scores on every benchmark it reports, faster task completion, shorter outputs. It is served under the existing deepseek-flash API model name at the same price. The legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work and are now routed to this checkpoint.

Price: $0.15 per 1M input tokens off-peak ($0.30 at peak), $0.60 per 1M output tokens, $0.003 per 1M cached input tokens. Flash input costs about one-fourth of DeepSeek V4 Pro, output about one-third. One independent 56-trial pilot measured $0.0157 per finished coding task, versus $0.0577 for GPT-6.1 Sol and $0.0653 for Claude Sonnet 5.5.

Architecture: Causal Encoder-Decoder sparse mixture-of-experts, 552B-parameter backbone plus 196B Engram parameters, 8B parameters active on input and 16B on output. 1M-token context window, up to 384K output tokens, native image understanding. FP4 KV cache at 890 bytes per token, roughly a quarter of the previous Flash generation.

Benchmarks (all DeepSeek-reported, not independently reproduced): DeepSWE v1.1 77.0 (September checkpoint 74.2, Claude Opus 5 74.0, GPT-5.6 Sol 73.0). Terminal-Bench 4.0 42.3 (was 31.2). Humanity's Last Exam without tools 42.6 (Opus 5: 56.3). DeepSeek's table shows it beating Kimi K3 on 10 of 11 tests and GPT-5.6 Sol on 9, trailing Opus 5 on 7.

Search & Fetch APIs: first-party web search and page-fetch endpoints for agents, released the same day. Each costs $0.75 per 1,000 successful calls, capped at 5 QPS per account. Calls through server-side tools in DeepSeek Harness, Claude Code, and Codex stay free for DeepSeek API users until per-request billing starts on November 1, 2026, Beijing time.

How to use: web and app free as before; API model name deepseek-flash; temperature 1.0, top_p 1.0 recommended. Tuned for DeepSeek Harness (dsh, v0.2.1-rc.1 or later recommended), Tencent CodeBuddy, and WorkBuddy. Weights MIT-licensed on Hugging Face, reported at 510 GB — check the model card before downloading. Companion technical report: "DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression".

Guide: https://lab.awaara.cn/deepseek-v4-1-flash-1017/
Related: GPT-6 Luna Pricing https://lab.awaara.cn/gpt-6-luna-pricing/
Related: Gemini 4 Argon https://lab.awaara.cn/gemini-4-argon/
Related: Muse Spark Open https://lab.awaara.cn/muse-spark-open/
Privacy: https://lab.awaara.cn/privacy/
Terms: https://lab.awaara.cn/terms/
All tools: https://lab.awaara.cn/