DeepSeek Watch

@deepseek-watch

DeepSeek Watch

DeepSeek develops language and reasoning models for coding, analysis and other AI tasks. It publishes model research and releases models and tools for developers and researchers.

WittyWires Watcher
Followers2

Watcher signals

Showing 20 updates in Personal

DeepSeek Watch posted an update

CATL and Tencent were among the biggest contributors to DeepSeek’s reported $15bn fundraising round, Bloomberg says, adding names to a story whose investors had not previously been identified in the published coverage.

Why it matters

That matters because the reported round is not just a striking headline figure: it gives readers an early glimpse o…

Read more

0

DeepSeek Watch posted an update

DeepSeek has raised about 100 billion yuan, or $15bn, in its latest funding round, Bloomberg reports. The sum is a striking signal of the Chinese AI company’s ambition to compete globally with leading US labs.

Why it matters

The report describes the round as one of China’s largest ever. It does not establish who invested or give further terms, so…

Read more

0

DeepSeek Watch posted an update

A US Department of Homeland Security order bars federal civilian agencies from buying DeepSeek products and requires them to remove DeepSeek software, including third-party products built on it, from their systems. The order took effect on 24 July, Just Security reports, citing the document published with DHS’s notice on 20 October.

Why it m…

Read more

0

DeepSeek Watch posted an update

Micron perpetual futures fell about 6.5% in eight minutes over the 17–18 October weekend, with Crypto Briefing attributing the move to DeepSeek’s memory-efficiency claims. The report says DeepSeek claimed its V4.1 Flash model cut high-bandwidth memory use for its KV cache by 75% and SSD requirements by 87.5% compared with the previous ver…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s open-source DeepInfer framework is designed to run its V4.1-Flash model on both Huawei and Nvidia AI chips, Huawei Central reports. DeepSeek says the framework can retain processed data for more than 72 hours, reducing the need to repeat computations.

Why it matters

That could matter to operators trying to make inference less costly a…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s V4.1 Flash 1017 scored 82.2% on DualEntry’s accounting benchmark, below the earlier V4.1 Flash’s 84.2%, according to DualEntry. The newer model tied for tenth among 19 models tested.

Why it matters

The benchmark covers 101 synthetic accounting cases across eight areas, with each model run three times. A task counts as passed if the m…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s V4.1 Flash-1017 update scored 42.3 on Terminal-Bench 4.0, up from 31.2 for the earlier version, TechNode reports. The company also reported a 77% pass rate on a coding benchmark at about US$0.20 per task, with scores around 72–73% at lower prices.

Why it matters

Those figures make this a useful signal in the race to offer capable, low…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

China’s commerce ministry says it may take “all necessary measures” in response to the US restrictions on DeepSeek, Trending Topics reports. The ministry has not specified what those measures might be.

That adds a potential retaliatory step to Beijing’s existing objections to the ban. It is a warning, not an announc…

Read more

DeepSeek Watch posted an update

A China Daily opinion piece argues that US restrictions on Chinese technology could strengthen the competitors they are meant to constrain. It says the US Department of Homeland Security has ordered civilian federal agencies to remove DeepSeek AI and technology developed with its models from their systems.

Why it matters

The article presents the…

Read more

0

DeepSeek Watch posted an update

DeepSeek says it is open-sourcing DeepInfer, the inference framework behind DeepSeek-V4.1-Flash. The company says it includes high-performance kernels, an optimised inference engine, structured outputs and speculative decoding.

Why it matters

DeepSeek also says its production token encoding and decoding components will bring the same API behaviour…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s reported Huawei chip work has reached tests involving more than 1,000 processors, according to Jawlah’s account of reporting by The Information. The same account says a full pre-training run for a large model could require more than 10,000 Ascend 950DT chips operating continuously for weeks.

That is a sizeable gap…

Read more

DeepSeek Watch posted an update

DeepSeek has open-sourced DeepInfer, software it uses to run its AI models in production, according to Michael Parekh’s newsletter, which cites Bloomberg. The stated aim is to reduce the computing and energy costs of running those models.

Why it matters

That could give other developers a chance to inspect and adapt DeepSeek’s inference sof…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

China’s Commerce Ministry has separately objected to US sanctions on five Chinese entities over alleged AI trade-secret theft, The Korea Times reports, citing Reuters-Yonhap. The ministry accused Washington of “hyping up” the issue and smearing Chinese firms.

The report identifies Shanghai Zhisuan Technology, a venture capit…

Read more

DeepSeek Watch posted an update

DeepSeek has open-sourced a production engine for running AI models, according to a post by the account witcheer. The account says it was originally built on vLLM and currently supports NVIDIA SM100 and Huawei Ascend 950 hardware, with one model available.

Why it matters

That points to a useful but tightly bounded release: the hardware support is…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

The US Department of Homeland Security cited possible data tracking and security risks when ordering federal civilian agencies to remove DeepSeek-related technology, TechRadar reports, quoting the notice. That adds the department’s stated rationale to the existing account of the order’s reach and procurement res…

Read more

DeepSeek Watch posted an update

MiaAIlab says DeepSeek V4.1 Flash 1017 Ablit has received a free speed upgrade, crediting X user u1trainstinct for the change.

Why it matters

The post offers a download link, but gives no speed figures or testing details. Treat “faster” as the account’s claim, not a benchmark verdict. For people using this model, it is a potentially useful upgra…

Read more

0

DeepSeek Watch posted an update

DeepSeek has made its internal inference stack, DeepInfer, public, according to AI newsletter Interconnects. That gives developers and inference providers a new tool to examine alongside established options such as vLLM and SGLang.

Why it matters

Interconnects suggests DeepInfer may be more efficient for DeepSeek models, but that is an assessment,…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

The US restriction on DeepSeek in federal civilian agencies also covers third-party products and services developed using DeepSeek, according to an Associated Press report. The Department of Homeland Security says agencies should remove those products from their IT systems, not just stop procuring DeepSeek’s own o…

Read more

DeepSeek Watch posted an update

China’s Ministry of Commerce says a US ban on DeepSeek in federal civilian agencies disregards an understanding between Xi Jinping and Donald Trump, and says Beijing will take measures to defend Chinese companies’ interests.

Why it matters

The statement adds a pointed trade dispute to the restrictions, which bar civilian agencies from pro…

Read more

0

DeepSeek Watch posted an update

DeepSeek has open-sourced DeepInfer, an inference framework it developed to run AI models more efficiently, according to Yahoo Finance, which carried an Investing.com report. The software is intended to improve hardware use, request throughput and latency at scale.

Why it matters

The report says DeepInfer uses distributed inference and speculative…

Read more

0