DeepSeek released DeepSeek-V4.1-Flash-1017, an update to its V4.1 Flash model, on October 17, 2026, and opened its Search and Fetch APIs to everyone at $0.75 per 1,000 calls for each endpoint. The update replaces V4.1 Flash on DeepSeek's website, app and API. According to DeepSeek's change log, "The API calling method and pricing remain unchanged," and developers use the deepseek-flash model name to reach the new version.

#What changed?

DeepSeek-V4.1-Flash-1017 is a point update to DeepSeek-V4.1-Flash, which DeepSeek first released on September 10, 2026. The Hugging Face model card calls it "a minor update to DeepSeek-V4.1-Flash, delivering substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks."

In its release note, DeepSeek claims "Stronger benchmark performance, faster task completion, and shorter outputs than the previous version." DeepSeek also says the model "uses reinforcement learning with self-feedback to improve end-to-end project development." DIGITIMES reported the same release, writing that DeepSeek released the update "on October 17, improving coding and reasoning performance while reducing task execution time and output length."

DeepSeek says the model is "Optimized for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy, and more," and recommends DeepSeek Harness v0.2.1-rc.1 or later.

#What do DeepSeek's own benchmarks show?

These results come from DeepSeek's own model card, not from an independent test. DeepSeek's figures for the update against the previous V4.1 Flash:

Benchmark (DeepSeek's own results)

V4.1 Flash 1017

V4.1 Flash

HLE (without tools)

42.6

36.8

Terminal-Bench 4.0

42.3

31.2

DeepSWE v1.1

77.0

74.2

ProgramBench

29.1

20.3

Agents' Last Exam

35.4

31.8

Apex Bench v1.1

72.1

64.6

DeepSeek's model card says it used the Claude Code harness for Terminal-Bench 4.0 and the Mini-SWE-V2 harness for DeepSWE v1.1. The same DeepSeek table lists Opus 5 at 51.8 on Terminal-Bench 4.0 and 74.0 on DeepSWE v1.1. By DeepSeek's own numbers, the update trails Opus 5 on Terminal-Bench 4.0 and edges it on DeepSWE v1.1.

#What are the Search and Fetch APIs?

DeepSeek opened two web tool endpoints, Search and Fetch, to all developers. The release note says they are "Powered by our in-house search engine, serving hundreds of millions of requests daily worldwide." In its change log, DeepSeek says its search engine "indexes hundreds of billions of web pages, supports multilingual search," and "filters SEO spam, GEO manipulation, and low-quality AIGC." These are DeepSeek's descriptions of its own product.

#How much does it cost?

DeepSeek says model pricing on deepseek-flash is unchanged. The Search and Fetch APIs are priced separately:

Item

Price, per DeepSeek

deepseek-flash API calls

Unchanged from V4.1 Flash

Search API

$0.75 per 1,000 calls

Fetch API

$0.75 per 1,000 calls

Search and Fetch via server-side tools in DeepSeek Harness, Claude Code and Codex

Free for DeepSeek API users until October 31, 2026, 16:00 UTC (21:30 IST)

DeepSeek's release note says these server-side tool calls "are currently free for DeepSeek API users" and that billing starts on October 31, 2026 at 16:00 UTC.

#Who can use it and when?

DeepSeek lists four ways to get the update, which it released on October 17, 2026.

Channel

Status

Source

DeepSeek website and app

Updated to the new model

Change log

DeepSeek API (deepseek-flash)

Updated, same calling method and price

Change log

Hugging Face weights

Available, MIT License

Model card

ModelScope weights

Available

Release note

The model card says: "This repository and the model weights are licensed under the MIT License." Hugging Face lists the model at 763B parameters with a 1M token context window.

#How to get it

  1. API: keep calling deepseek-flash; DeepSeek says it now points to the latest model.
  2. Harness users: upgrade to DeepSeek Harness v0.2.1-rc.1 or later, as DeepSeek recommends.
  3. Open weights: download from Hugging Face or ModelScope. The model card notes: "This release does not include a Jinja-format chat template." DeepSeek instead ships a Python reference encoder and deepseek-recipe, "a set of Rust libraries with Python bindings."

#How does it compare?

The update keeps DeepSeek's price on deepseek-flash while adding an agent web tool with a published per-call price. For a recent launch on a different pricing model, see our coverage of Meta's Muse Wonder and Muse Wonder Full. The free Search and Fetch period applies inside Claude Code and Codex; our comparison of OpenAI Dots, ChatGPT agent and Codex covers Codex, and our guide to controlling what AI coding agents can reach on a Mac covers the network side of agent web access. More developer coverage is in the Developers hub.

#What is still unclear?

  • Independent results. All benchmark figures are DeepSeek's own; no independent evaluation is in the sources.
  • Billing details. DeepSeek gives a per-call price but the sources do not list rate limits for Search and Fetch.
  • Model pricing. DeepSeek says pricing is unchanged but the sources reviewed do not restate the per-token rates.