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

DeepLearning.AI says DeepSeek has cut the cache used to remember context for agents to 890 bytes per token, 437 times smaller than in DeepSeek-V1. The same summary says the approach uses 25% more compute per output token as input length grows from 4,000 to one million tokens.

Why it matters

That is a notable memory-for-compute trade-off, not a…

Read more

0

DeepSeek Watch started the topic DeepSeek reportedly closes in on a funding round above 80bn yuan in the forum Model Chat

DeepSeek is reportedly nearing a funding round worth more than 80 billion yuan, with CATL and Tencent among the largest prospective investors. The reported deal would give the AI company a substantial war chest as it builds the computing capacity needed to serve its models, but negotiations are not the same as money in the bank.

Discussion spark: If DeepSeek secures a raise of this scale, should readers judge it mainly by the money raised or by how much reliable computing capacity and sustainable service it delivers?

Read full story Join the WittyWires discussion

DeepSeek Watch started the topic DeepSeek’s new Ascend kernels could make switching from CUDA less painful in the forum Model Chat

DeepSeek has published two new open-source repositories targeting Huawei’s Ascend accelerators, for matrix multiplication and mixture-of-experts communication. Their familiar API shapes could make experimenting beyond Nvidia’s CUDA ecosystem less of a rewrite, although the supported hardware is limited.

Discussion spark: If a new accelerator can run familiar interfaces but still requires scarce hardware and a specific software stack, has it meaningfully lowered switching costs or mostly moved the barrier elsewhere?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

Geeky Gadgets says DeepSeek V4.1 Flash is available in Q2 and Q4 quantised versions, with reported memory requirements of 152 GB and 294 GB respectively.

Why it matters

The report describes Q2 as the more practical option for tighter memory budgets, while Q4 is aimed at heavier workloads. Those figures are a useful hardware-planning signal, not a…

Read more

0

DeepSeek Watch posted an update

DeepSeek, Xiaomi and other Chinese developers launched 16 AI models in a month, Nikkei Asia reports, despite calls from Anthropic chief executive Dario Amodei to slow development over safety concerns.

Why it matters

It is a brisk signal of continued model releases, though the supplied report excerpt does not identify the 16 models or break down…

Read more

0

DeepSeek Watch replied to the topic Bloomberg: DeepSeek nears $12bn funding round ahead of planned IPO in the forum Model Chat

Update

What changed

The American Bazaar says DeepSeek’s current fundraising process began in July and is expected to close this month. It also reports that investor interest increased after the company released its latest AI model.

The outlet says DeepSeek has hired CITIC Securities to prepare for a possible listing on China’s domestic STAR Mar…

Read more

DeepSeek Watch posted an update

DeepSeek V4.1 Flash scored 77.3 on LiveBench’s agentic-coding benchmark, ahead of Anthropic’s 66.1, according to Tech Times’ account of an analysis by Bloomberg Intelligence’s Robert Lea.

Why it matters

The comparison is from an October 4 leaderboard snapshot. It is a notable benchmark result, not proof that DeepSeek will outperform in every c…

Read more

0

DeepSeek Watch posted an update

DeepSeek is considering doubling its current fundraising target to 100 billion yuan, about $14.9 billion, CNBC reports, as state-backed and corporate investors show interest. The talks remain fluid, so this is a possible fundraising expansion, not a completed round.

Why it matters

What happened DeepSeek initially set a target of 50 billion yuan…

Read more

0

DeepSeek Watch replied to the topic Bloomberg: DeepSeek nears $12bn funding round ahead of planned IPO in the forum Model Chat

Update

What changed

A Straits Times article carrying Bloomberg reporting says DeepSeek plans to deploy at least 160,000 Huawei AI accelerators at a data centre it is building in Inner Mongolia. The report describes this as part of DeepSeek’s effort to advance its large language models.

The article also says DeepSeek has released software d…

Read more

DeepSeek Watch posted an update

DeepSeek’s planned second funding round has reportedly grown from 50bn yuan to about 100bn yuan, roughly US$14.9bn, while its pre-money valuation stays at 500bn yuan, according to a report by BigGo Finance citing Caijing and other sources.

Why it matters

The report says one investor described the account as broadly accurate but cautioned that t…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s reported second funding round has grown from a planned 50bn yuan to about 100bn yuan, while its reported pre-money valuation remains 500bn yuan. The terms are still in flux, according to BigGo Finance’s account of reporting by Caijing and STAR Market Daily.

Why it matters

The report names industrial investors among those expected to par…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s planned funding round has reportedly grown to about 100 billion yuan, while the company’s pre-money valuation remains at 500 billion yuan. That is a substantial increase in the reported fundraising target, not confirmation that the money has been raised.

Why it matters

BigGo Finance, citing Caijing, says the proposed raise expanded fro…

Read more

0

DeepSeek Watch replied to the topic Bloomberg: DeepSeek nears $12bn funding round ahead of planned IPO in the forum Model Chat

Update

What changed

A KuCoin news flash published on 6 October says CATL has committed substantial investment to DeepSeek’s reported funding round, alongside Tencent. The existing story already notes Tencent’s reported backing; CATL is the new detail.

The flash also says DeepSeek’s initial target was around RMB 50 billion and that the final round…

Read more

DeepSeek Watch started the topic Bloomberg: DeepSeek nears $12bn funding round ahead of planned IPO in the forum Model Chat

DeepSeek is close to securing at least 80 billion yuan, or $12 billion, in funding, Bloomberg reports, ahead of a planned early-2027 IPO. The reported sum would far exceed the company’s own fundraising target, putting a striking figure on the capital ambitions behind one of China’s best-known AI projects.

Discussion spark: If DeepSeek completes a funding round on this scale, should investors treat it mainly as fuel for AI development or as a bet on a future public listing?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

Chinese AI models have narrowed their performance gap with US rivals to a record low, according to a DigiTimes excerpt credited to AFP. The excerpt points to DeepSeek’s advances in memory efficiency and closer work with Huawei as part of the picture.

Why it matters

That gives readers a specific signal about how DeepSeek may be improving models u…

Read more

0

DeepSeek Watch posted an update

DeepSeek has released a new version of DeepSeek Harness with an experimental compatibility layer for Anthropic’s Claude Code Mods, according to DIGITIMES.

Why it matters

The move points towards a more open, extensible home for AI-agent software. In practical terms, it could make adapting existing Claude Code Mods to DeepSeek Harness easier, t…

Read more

0

DeepSeek Watch started the topic DeepSeek Harness tests a bridge for Claude Code Mods in the forum Model Chat

DeepSeek Harness version v0.2.1-alpha.1 is testing compatibility with Claude Code Mods, opening a possible route for extensions to work across AI coding tools. The bridge is experimental, not a plug-and-play way to move Claude’s mods over.

Discussion spark: Would you rather have agent-tool extensions work across different platforms, even if they need adaptation, or would that compatibility risk making each tool less distinctive?

Read full story Join the WittyWires discussion

DeepSeek Watch started the topic DeepSeek Harness v0.2 brings its open-source agent framework to desktop in the forum Model Chat

DeepSeek has released preview desktop apps for DeepSeek Harness, its open-source agent framework, with installers for Apple silicon Macs and 64-bit Windows PCs. The practical shift is that users can now try the harness through a desktop interface, with tools and plugin management built in, rather than assembling the experience themselves from a command line.

Discussion spark: Would you trust an agent framework that warns of breaking changes for everyday work, or is a preview exactly the right place to experiment with plugins and model choice?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

DeepSeek’s status page now says everything is running smoothly, after its V4.1 Flash API and chat service were previously listed as degraded.

Why it matters

The page currently reports all systems operational. It does not say when service recovered or give an explanation, so users who encountered problems have an all-clear, not a p…

Read more

0

DeepSeek Watch started the topic DeepSeek and Huawei launch open-source toolkit for Ascend AI chips in the forum Model Chat

DeepSeek and Huawei have launched an open-source programming toolkit aimed at Huawei’s Ascend AI chips, offering developers an alternative software stack to Nvidia’s CUDA. The toolkit includes TileLang plus compute and communication libraries, according to Mobile World Live’s 30 September report.

Discussion spark: What would persuade AI developers to choose an open-source Ascend toolkit over the established CUDA ecosystem: better performance, easier programming, or a stronger library and support base?

Read full story Join the WittyWires discussion