DeepSeek raises API pricing up to 1,100%, reopens $8bn round, invests in robotics

DeepSeek Abandons Cheap-AI Strategy with Dramatic Price Increases

DeepSeek, the Chinese AI startup that built its reputation on affordability, is implementing dramatic API pricing increases ranging from 50% to over 1,100% for its V4-Pro model and V4-Flash model, effective August 16 at 16:00 UTC1

1

. The new pricing structure replaces the current flat rate with peak and off-peak billing, where peak hours run from 01:00-04:00 and 06:00-10:00 UTC, with off-peak rates set at half the peak rates5

5

.

For V4-Flash, output tokens will jump to $1.32 per million during peak hours and $0.66 per million during off-peak hours, up from the current flat rate of $0.28 per million5

5

. V4-Pro output tokens will rise even more steeply to $3.96 per million at peak and $1.98 per million off-peak, compared with the current $0.87 per million. Input token prices are climbing similarly, with V4-Flash cache-miss input tokens rising to $0.44 per million at peak hours from $0.14, while V4-Pro cache-miss input tokens will increase to $1.32 per million at peak from $0.4355

5

. The company stated it is adjusting API pricing "to allocate resources more reasonably," aiming to shift developer workloads toward less congested periods for better resource allocation5

5

.

$8 Billion Funding Round Signals Major Strategic Shift

DeepSeek has reopened its second funding round, seeking close to $8bn at a valuation of approximately 500 billion yuan, or $74bn2

2

. Monolith Management, an early backer of Chinese AI champion Moonshot, is in talks to join the round. This represents a significant increase from the startup's first external funding round earlier this summer, which closed near 350 billion yuan and roughly doubled founder Liang Wenfeng's net worth to about $36bn2

2

.

The round is technically a restart rather than a fresh start, as DeepSeek paused the process last month after leaked remarks from Liang to investors caused friction2

2

. The timing and investor list can still change, but the intended use of funds marks a fundamental departure from the company's efficiency-focused origins. DeepSeek plans to spend part of the capital building its own data centres, led by a large facility in Inner Mongolia2

2

. The lab that made its name on efficiency now needs its own compute infrastructure, signaling that the next phase of AI development will require substantial capital investment.

Strategic Investment in Unitree Robotics for Embodied Intelligence

DeepSeek invested 140.8 million yuan, approximately $20.8m, in Unitree Robotics' Shanghai IPO, receiving 933,399 shares representing 2.31% of the strategic placement with a 36-month lock-up period3

3

. The two Hangzhou-based companies agreed to jointly develop AI models for humanoid robots, combining DeepSeek's expertise in AI models with Unitree's work in mechanical engineering, motion control, and embodied intelligence3

3

.

Under the partnership, Unitree would give preference to DeepSeek when procuring model-training services and technical solutions, while DeepSeek would similarly favor Unitree when purchasing humanoid robots or exploring embodied-AI applications3

3

. This collaboration targets a critical bottleneck for humanoid developers: building a robot "brain" capable of understanding unfamiliar surroundings and reliably turning instructions into physical actions. While DeepSeek's AI models have performed strongly on coding, mathematics, and reasoning tests, the company's strengths have been concentrated in language-model series rather than multimodal systems able to process vision alongside text3

3

.

Unitree priced its IPO at 150.80 yuan per share, raising about 6.1 billion yuan, or $904m, at a valuation near $9bn, making it the first humanoid-robot maker to list on the mainland2

2

. The offering carries a price-to-earnings ratio of 219, significantly higher than the industry average of near 392

2

. Backers already include vehicles tied to Tencent, Alibaba, and China Mobile. Revenue more than quadrupled to 1.7 billion yuan in 2025, with humanoids becoming the biggest product line, though first-quarter profit fell 52.6% as research and marketing costs climbed2

2

.

DeepSeek Harness Challenges Anthropic's Claude Code

DeepSeek launched DeepSeek Harness v0.1, an open-source agent harness that provides developers an alternative to integrated coding-agent environments such as Anthropic's Claude Code4

4

. Built on the Cordis framework, Harness operates on an unusually modular premise where practically every component of the agent runtime can be swapped out as a plugin4

4

.

Available under the MIT license on GitHub, DeepSeek Harness treats everything as a plugin, extending to models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and user interfaces4

4

. Rather than making these components fixed pieces of a single coding agent, Harness lets developers mix, replace, and extend them. The repository garnered roughly 27,500 GitHub stars and 2,000 forks on launch day, though the project is explicitly described as a developer preview with warnings that "THERE WILL BE COMPATIBILITY-BREAKING CHANGES"4

4

.

Harness can already inspect repositories, edit files, execute shell commands, search files and the web, maintain plans, invoke skills, delegate work to subagents, and enforce approval policies—essential capabilities for agentic workloads that distinguish it from autocomplete systems4

4

. DeepSeek describes Standard mode as a full coding agent with file editing, shell access, search, planning, subagents, and workflows, with a local web interface that lets users select a workspace and approve sensitive operations.

V4-Pro Receives Agent-Focused Enhancement

Alongside Harness, DeepSeek released the general-availability version of DeepSeek-V4-Pro-0813, an updated flagship model focused heavily on agentic workloads4

4

. V4-Pro is now available across DeepSeek's web interface, mobile app, and API, with native support for the OpenAI Responses API and integration with Codex. The 1.6-trillion-parameter V4-Pro model activates 49 billion parameters per token, while the smaller 284-billion-parameter V4-Flash activates 13 billion, both supporting context windows of up to one million tokens4

4

.

Despite the substantial price increases, DeepSeek's new peak rates remain below some competitors, with Anthropic's Fable 5 charging $50 per million output tokens5

5

. DeepSeek released V4-Flash earlier this month, with research firm Artificial Analysis finding it was the least expensive well-known AI model to run globally at the time, costing roughly 3 cents per benchmark test5

5

. The Hangzhou-based startup had previously offered a 75% promotional discount on V4-Pro through May 5, underscoring how dramatically its pricing strategy has shifted in recent months. The company has also begun laying groundwork for an IPO alongside closing its first outside funding round5

5

.