DeepSeek Secures $97B Funding, Launches Code Initiative, and Locks in Permanent 75% API Discount

This week DeepSeek released a cluster of announcements: a 700 billion RMB (~$97 billion) financing round, the upcoming DeepSeek Code product, the community‑driven DeepSeek‑TUI project reaching 33.5 k stars, and a permanent 75% discount on the DeepSeek‑V4‑Pro API.

Business Times reported that DeepSeek is pursuing a 700 billion RMB financing round, with state‑backed capital participation, underscoring strong institutional interest. Founder Liang Wenfeng stated that AGI is the core goal. The capital would expand compute resources and provide long‑term policy support, reducing pressure compared with purely market‑driven firms. Although valuation details remain unconfirmed, the amount would place DeepSeek in the same scale as OpenAI and Anthropic.

DeepSeek Code was hinted at in a recruitment post by employee Deli Chen, who announced the creation of a new Harness team in Beijing. Two positions were opened: a Harness product manager responsible for product planning and a Harness R&D engineer handling low‑level implementation. Harness is the execution layer of AI coding agents, used by Claude Code, and DeepSeek aims to build its own coding‑agent product to compete with Cursor and Claude Code. The team is in early hiring, and the main challenges lie in product engineering—context management, tool‑call stability, and multi‑file collaboration.

The open‑source project DeepSeek‑TUI, started by US developer Hunter Bown, grew from 1.4 k stars in early May to 33.5 k stars in just over three weeks—a twenty‑fold increase. Its core capabilities include an Auto mode that switches between Flash and Pro inference, 1 M‑token context with prefix cache, three operation modes (Plan, Agent, YOLO), workspace rollback via side‑git snapshots, and LSP diagnostics integrated into the TUI. Implemented in Rust, it supports DeepSeek’s official API, NVIDIA NIM, OpenRouter, Ollama, and self‑hosted SGLang/vLLM, making it friendly for users who run their own inference services. Installation is simple: npm install -g deepseek-tui DeepSeek also announced that the 75% discount for DeepSeek‑V4‑Pro (input cache‑hit 0.0145, input cache‑miss 1.74, output 3.48) is now permanent, removing the previous limited‑time renewal. This gives developers a stable cost baseline for project budgeting.

The convergence of a massive financing round, the formation of a dedicated Code team, a thriving community project, and a locked‑in low‑price API suggests a coordinated strategy: funding fuels compute and long‑term R&D, the Code product fills the developer‑tool gap, and the permanent discount secures ecosystem adoption. When DeepSeek Code launches, comparing it with Cursor and Claude Code will provide a clear view of its competitiveness in the increasingly brutal coding‑agent market.