DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, 2026, an update to its smallest V4.1 model that now serves the deepseek-flash API name, the DeepSeek website and the app. On the same day it opened Search and Fetch, two paid APIs built on the in-house search engine behind its own chatbot. API prices for the model did not change, and the weights are on Hugging Face.

Key Facts

Details
Model DeepSeek-V4.1-Flash-1017, replacing the September 10 release of V4.1 Flash
API name deepseek-flash; the retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp are served by 1017 at Flash prices
Model pricing Unchanged. Peak rates are $0.30 per million input tokens (cache miss) and $1.20 per million output tokens; off-peak rates are half
Context and output 1M-token context, up to 384K output tokens
Weights deepseek-ai/DeepSeek-V4.1-Flash-1017 on Hugging Face
New APIs POST https://api.deepseek.com/v1/tools/search and POST https://api.deepseek.com/v1/tools/fetch
Search and Fetch price $0.75 per 1,000 successful requests each; failed calls, empty results and timeouts are not billed
Rate limit 5 requests per second per account for each API

Benchmark Changes

DeepSeek's changelog lists scores for the September release and for 1017. On the benchmarks that appear in both lists, every score went up:

Benchmark V4.1 Flash (Sep 10) V4.1-Flash-1017
HLE, no tools 36.8 42.6
Terminal-Bench 4.0 31.2 42.3
DeepSWE v1.1 74.2 77.0
ProgramBench 20.3 29.1
NL2Repo-Bench 65.4 68.2
Agents' Last Exam 31.8 35.4
Chartography, with tools 78.9 83.0
BabyVision, with tools 89.6 92.5

DeepSeek ran Terminal-Bench 4.0 with Claude Code as the agent framework. The company also says 1017 finishes tasks faster and writes shorter outputs. It tuned the model for its own DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, and recommends Harness 0.2.1-rc.1 or later.

The Search and Fetch APIs

Search takes a query and returns ranked web results. Fetch takes a single URL and returns the page as Markdown. Both use the same API key as DeepSeek's models. DeepSeek describes the engine behind them as its own general-purpose search engine, which it says indexes hundreds of billions of pages and handles hundreds of millions of requests a day.

Developers who use the DeepSeek API through DeepSeek Harness, Claude Code or Codex already hit these APIs through server-side tool calls. Those calls are free for now. Billing per request starts at 00:00 Beijing time on November 1, 2026, which is 16:00 UTC on October 31.

Reactions

Arena said on October 18 that DeepSeek-V4.1-Flash-20261017 (Max) had become the top open model on its Code Arena WebDev leaderboard, ranking 12th overall with 1,670 points and gaining 52 points over the September version. The Decoder noted that independent benchmarks of the model had not yet been published when DeepSeek released it.

What We Don't Know Yet

  • How the model does in independent evaluations other than Arena's leaderboard. The scores above come from DeepSeek.
  • Whether direct Search and Fetch calls made outside Harness, Claude Code and Codex get any free period. The pricing page only exempts those three routes.
  • How long the retired V4 Flash model names will keep routing to the new model. DeepSeek calls the routing temporary.

Sources