DeepSeek Updates V4.1 Flash and Opens Search API

DeepSeek has updated its V4.1 Flash model, saying the October 17 release can finish coding and agent tasks faster and with shorter outputs than the previous version. The model, named DeepSeek-V4.1-Flash-1017, is already in the company’s web app, mobile app and API. This is a refresh of V4.1 Flash, not the launch of a new V4 Pro flagship.

For developers, the practical change is unusually simple. DeepSeek says the API name remains deepseek-flash and its price has not changed. Existing integrations using that name should receive the updated model without switching endpoints. DeepSeek also opened access to its own Search and Fetch API on the same day, giving applications a way to retrieve current web information through DeepSeek’s search system.

What has improved?

DeepSeek’s official changelog reports a score of 42.3 on Terminal-Bench 4.0, compared with 31.2 for the September V4.1 Flash release. It lists 77.0 on DeepSWE v1.1, up from 74.2, and 29.1 on ProgramBench, up from 20.3. Those are DeepSeek’s published test results, not independent proof that every user will see the same improvement. The company used specific testing frameworks, including Claude Code for Terminal-Bench 4.0 and Mini-SWE-V2 for DeepSWE.

The less glamorous claim may matter more in production. DeepSeek says it has cut task runtime and output length while improving results. An agent that reaches an answer with fewer steps or tokens can cost less to operate even when the price per token stays the same. That is why total cost matters more than a new discount on the API price. Real savings will depend on the task, the software around the model and how often a run has to be retried.

The company says it optimised the refresh for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy and other agent frameworks. It recommends DeepSeek Harness 0.2.1-rc.1 or later for the best fit with the new model. The updated model is also listed on DeepSeek’s Hugging Face page, giving developers another route to examine the release rather than relying only on the hosted API.

DeepSeek’s new search API sits alongside the model update. The company says its in-house engine indexes hundreds of billions of pages, supports multiple languages and filters low-quality AI-generated material and search spam. Those scale and quality claims come from DeepSeek and need independent testing. Still, the combination matters: developers can now pair a cheaper, faster agent workflow with current information from a first-party search service.

It is a significant step in a market where AI providers increasingly compete on the complete task, not just the model’s raw score. A useful agent needs to find information, reason over it, use tools and return a reliable answer at a price a business can sustain. Search also raises familiar questions about source quality, attribution and whether retrieved pages actually support an AI-generated response.

TechBooky has tracked DeepSeek’s pressure on AI agent pricing since the earlier V4 Flash API release. This update sharpens that argument without changing the sticker price. The next test is whether independent users see the same speed and coding gains on ordinary workloads, and whether DeepSeek’s search API consistently brings back dependable sources. For now, developers have a newly updated Flash model to try, while readers should treat the benchmark figures as company claims until broader testing catches up.