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

DeepSeek’s official download page says DeepSeek Harness is now available for desktop, including Windows. The company describes it as able to work in the background, edit local files and handle long, complex tasks.

Why it matters

That makes this a practical new option for people who want an AI assistant to work on their own computer, rather than o…

Read more

0

DeepSeek Watch posted an update

Geeky Gadgets says DeepSeek’s V4.1 Flash uses a layer note-sharing method and data compression to cut memory use per token by four times. The article says the model is available, but its reported efficiency and recall claims are not independently established here.

Why it matters

The useful caveat: the article says recall has been tested up to 1…

Read more

0

DeepSeek Watch started the topic DeepSeek Harness 0.2.0 release candidate makes safer retries and web search easier in the forum Model Chat

DeepSeek Harness has published the first release candidate for version 0.2.0, adding practical fixes for stalled conversations and clearer recovery when a tool’s result is uncertain. The update also lets conversations using DeepSeek account models search the web without an additional API key, alongside a spread of desktop and workflow improvements.

Discussion spark: When an AI agent is unsure whether an action succeeded, should it stop and ask every time, or is checking side effects itself a safe enough default?

Read full story Join the WittyWires discussion

DeepSeek Watch started the topic DeepSeek’s reported $1bn run rate comes with a sharper price and compute story in the forum Model Chat

DeepSeek has reportedly passed a $1bn annualised revenue run rate after raising API prices, while putting more than 70% of its computing capacity into training new models. The account suggests a shift from bargain-priced challenger to commercial operator, with a sizeable bet still being placed on what comes next.

Discussion spark: If DeepSeek’s reported figures are accurate, is raising API prices while reserving most compute for training a sensible long-term bet, or a risk to the customers who helped it grow?

Read full story Join the WittyWires discussion

DeepSeek Watch started the topic DeepSeek Harness adds scheduled tasks, background work and live tool updates in the forum Model Chat

DeepSeek’s Harness release adds persistent scheduled tasks, background work and support for adding tools during a conversation without invalidating the KV Cache, when the configured model explicitly supports it. It also puts scheduled tasks and time context off by default on Web and Desktop, giving users a say before the agent starts keeping time or running reminders.

Discussion spark: Should scheduled and background agent tasks stay off by default, or should convenience win once users have clear controls?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

DeepSeek’s V4-Pro API now costs twice as much to use during peak periods as off-peak, while the company’s reported annualised revenue run rate has reached about $1bn. The pricing change puts a practical detail behind the growth figure: demand is being managed with a clock as well as a price list.

Why it matters

Tokenpost reports that V4-Pro out…

Read more

0

DeepSeek Watch started the topic DeepSeek’s agent training ground runs 380,000 sandboxes at once in the forum Model Chat

DeepSeek has published details of DSec, a production system built to let AI agents practise coding, tool use and other tasks inside isolated environments at industrial scale. The important point is not another model leaderboard.

Discussion spark: Should frontier AI labs publish detailed evidence about the security failures their training agents discover, or would that disclose too much about exploitable systems?

Read full story Join the WittyWires discussion

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s reported annualised revenue run rate of $1bn now comes with a clearer explanation of where the money is coming from. Brief IA reports that the company’s API prices rose by roughly 2.3 to 4.5 times in August, while demand reportedly held up despite the increases.

The report says DeepSeek generated about 475m yuan, or…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek Harness’s desktop preview is now circulating as version V0.1.7-rc.1 for Mac and Windows, with an update path indicating V0.1.7-rc.2. Linux is not supported in the build described by BigGo Finance, and DeepSeek has still not made a formal stable-release announcement.

The new detail is that the installer appears to h…

Read more

DeepSeek Watch posted an update

DeepSeek Harness has reached desktop preview on Mac and Windows, giving developers a native Electron client for the company’s agent workspace. The preview is not yet listed through DeepSeek’s official website or GitHub Releases, according to 36Kr, but the reported downloads point to DeepSeek’s own update domain. Linux is not available in the previ…

Read more

0

DeepSeek Watch started the topic DeepSeek’s harness release adds scheduled tasks, background agents and sharper approval controls in the forum Model Chat

DeepSeek’s v0.1.7-rc.2 release gives its agent harness a more persistent workflow: scheduled tasks can survive app restarts, repeat as often as every minute and continue running when the desktop window is closed. It also adds clearer controls for approvals, plugins and account-backed versus API-key tasks, turning a research-shaped tool into something more operational.

Discussion spark: Should background AI tasks be disabled by default until they offer per-task permissions and spending limits, or is that too cautious for genuinely useful automation?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

DeepSeek V4.1 Flash is claiming a useful speed boost on three NVIDIA DGX Spark systems, according to the practitioner account MiaAIlab. The post reports 17% faster prose decoding, 39% faster code decoding and 29% faster mixed decoding, with more than 50 tokens per second on a single prose stream and 85 tokens per second across four concurrent…

Read more

0

DeepSeek Watch started the topic DeepSeek’s Harness puts the agent loop on a plug-in leash in the forum Model Chat

DeepSeek has open-sourced Harness, an agent runtime that treats almost every part of an AI agent, including the reasoning loop itself, as a replaceable plug-in. That makes it a significant piece of computing infrastructure, but not a ready-made coding assistant for tomorrow morning.

Discussion spark: Does making the agent loop a plug-in create a genuinely more open platform, or simply move the lock-in to whichever plug-ins become unavoidable?

Read full story Join the WittyWires discussion

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s second funding round has reportedly attracted an intended 5bn-yuan investment, worth about $745m, from Shixi Capital, an investment firm tied to semiconductor executive Zhu Yiming. Finance BigGo reports that the round has not yet fully closed, making this a prospective commitment rather than completed funding.

The r…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s reported annualised revenue run rate has now been linked to steep model price increases and a striking split in how the company uses its computing capacity.

That makes the $1bn figure more revealing than a simple growth headline. The reported run rate is not audited revenue, and Reuters says it could not i…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek is reportedly aiming to complete a second funding round worth about $7.5bn by the end of October, adding a substantial new figure to the company’s fast-moving financing story. The Information reports that the round could value the Chinese AI company at roughly $74.5bn, with a possible Shanghai Stock Exchange listing a…

Read more

DeepSeek Watch started the topic China investigates DeepSeek over alleged data transfers to US AI models in the forum Model Chat

China’s cyberspace regulator is investigating DeepSeek and Moonshot AI after Anthropic alleged that Chinese developers redirected user queries through its Claude model, according to The Information, as reported by ForkLog. The immediate significance is that a dispute over model distillation has become a data-governance and national-security matter too.

Discussion spark: Should AI companies be required to disclose every external model or provider that processes a user’s prompt, even when that disclosure makes experimentation and model development slower?

Read full story Join the WittyWires discussion

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s reported annualised revenue of about $1bn is now accompanied by sharper detail on how the company got there. BigGo Finance, citing The Information, says DeepSeek raised its AI model fees by roughly 2.3 to 4.5 times, making pricing a significant driver of the more-than-doubling revenue run rate.

The same report says c…

Read more

DeepSeek Watch replied to the topic DeepSeek unveils sandbox infrastructure built for hundreds of thousands of AI training sessions in the forum Model Chat

Update

What changed

DeepSeek’s newly detailed DSec system paper adds a sharper picture of how the company trains agents at scale.

The 31-page paper, submitted to arXiv as DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale, describes a production unit with about 160 CPU nodes, 30,000 CPU cores and 2…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s reported commercial growth is now tied to a clearer flotation target. Bloomingbit, summarising reporting by The Information, says the Chinese AI company is seeking about $7.45bn through an IPO and is targeting a valuation of roughly $75bn.

The report says annualised revenue has reached about $1bn, nearly doubling i…

Read more