News Date: 2026-10-24
October has been a month of rapid acceleration for DeepSeek. On one side, the long-rumored massive funding round has finally materialized. On the other, a new model update is rewriting benchmark records at an extremely low cost. For developers and enterprise users, DeepSeek is redefining AI’s cost-effectiveness with "less money, fewer tokens, and shorter time."
$15 Billion Funding: CATL and Tencent Lead the Charge
According to people familiar with the matter cited by Bloomberg, DeepSeek has raised approximately 100 billion yuan (about $14.98 billion) in its latest funding round[citation:5]. Battery giant CATL and WeChat operator Tencent Holdings were among the biggest investors in this round.
The investment list is noteworthy. Tencent and CATL were the two largest contributors. Other industrial capital, including Geely, also participated. The funds will primarily be directed toward model training, compute infrastructure, and talent acquisition, with a large data center in Inner Mongolia being a top priority.
V4.1-Flash-1017: Better Performance at a Lower Price
On October 17, DeepSeek quietly rolled out the V4.1-Flash-1017 version, with its official web, app, and API models all switching to the new iteration simultaneously[citation:3][citation:6]. While the version number suggests a minor update, the official benchmark data is quite compelling: on the DeepSWE v1.1 software engineering benchmark, the new model’s pass rate reached 77.0%, up from roughly 74% in the previous version[citation:3][citation:8].
More critically, the cost. According to official performance-cost charts, the V4.1-Flash-1017 achieves its highest pass rate at an average cost of only about 1.3 yuan per task[citation:13]. By comparison, many overseas models with similar performance require tens of yuan or more per task. DeepSeek states that the new model not only improves performance but also shortens task runtime and output length, allowing users to complete tasks at lower cost and higher speed.
"Wish-Driven Programming" Becomes Reality
The most intuitive change is the model’s ability to generate complete projects from scratch. In official demos, V4.1-Flash-1017, running within the DeepSeek Harness framework, completed tasks like 3D modeling of the Forbidden City, developing a web-based racing game, and even recreating scenes from *Portal 2* using the Godot engine, all with minimal interaction rounds[citation:8].
DeepSeek calls this capability "Wish-Driven Programming." Users simply describe what they want, and the model handles the entire task chain: understanding requirements, planning, code generation, runtime checks, and bug fixing. While this doesn’t mean the model replaces an entire dev team, it is evolving from a "coding assistant" into an "agent that can push a project to a runnable state."
Open Search Capabilities: Search & Fetch APIs
Alongside the model update, DeepSeek opened up its in-house search engine with Search and Fetch interfaces. Search is for retrieving web pages, and Fetch is for scraping content from specified URLs. Both are priced at 5 yuan per 1,000 successful requests, and billing only occurs when a request is successful and produces a valid result[citation:8][citation:13].
From a $15 billion funding round to ultra-low-cost models, from wish-driven programming to open search capabilities, DeepSeek’s October moves all point in the same direction: driving down the cost of AI usage while improving task delivery completeness. For the price-sensitive developer market and enterprise users needing large-scale deployment, this may be more compelling than pure benchmark improvements.