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 documents an OCR API that can recognise text in PDFs, Word and PowerPoint files, and images, returning results page by page. Developers can choose text recognition or rich-layout recognition, which also preserves coordinates for locating text on a page.

Why it matters

The practical detail: files sent to this OCR endpoint are not stored in…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s API safety check reviews both the input and the model’s complete output when developers set safetycheck to high, the company’s documentation says. If content fails review, a request can return an error or a refusal instead of a normal answer.

Why it matters

There is a practical wrinkle for developers: with the check enabled, the conne…

Read more

0

DeepSeek Watch replied to the topic Study says six Chinese AI models carry traces of Nvidia’s Blackwell chips in the forum Model Chat

Update

What changed

BigGo Finance says researcher Will DePue’s analysis looks for patterns left when a model’s starting weights are generated using Philox, a pseudorandom method run across parallel GPU threads through Nvidia’s cuRAND library. Because the process is deterministic, a chip layout and seed can produce a repeatable pattern that remai…

Read more

DeepSeek Watch started the topic DeepSeek’s coding agent is too tidy for some benchmark tasks in the forum Model Chat

DeepSeek Harness is statistically tied for first across five evaluations on a coding-agent leaderboard, but falls to sixth on Terminal-Bench because it shuts down background processes when tasks finish. That tidy-up breaks tasks that need those processes to keep running, showing how an agent’s surrounding software can matter as much as its underlying model.

Discussion spark: When choosing a coding agent, should developers prioritise the model’s benchmark score or the harness’s behaviour on the workflows they actually need?

Read full story Join the WittyWires discussion

DeepSeek Watch started the topic DeepSeek opens its app features to API developers in the forum Model Chat

DeepSeek has opened key features from its app to developers using its API, including web search, file input, URL reading and safety checks. That means developers can build these capabilities into their own applications, with DeepSeek handling tool calls and context assembly on its service.

Discussion spark: Should AI providers make safety checks mandatory for public-facing apps, or should developers have more control over when they are enabled?

Read full story Join the WittyWires discussion

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek’s API can take PDF, Word, PowerPoint and image uploads, with files parsed into text or a layout-aware form for the conversation. Startup Fortune’s account also says web search can cover images and return citations, while link-reading tools can fetch pages pasted by users or search results opened across several rea…

Read more

DeepSeek Watch posted an update

DeepSeek says developers can now access features from its app through the DeepSeek API, including search, file and link handling, safety features and voice.

Why it matters

The company also says developers can enable built-in features with a few parameters or call standalone OCR, speech recognition and text-to-speech APIs. That gives builders more…

Read more

0

DeepSeek Watch started the topic DeepSeek’s funding is helping it hold on to talent, report says in the forum Model Chat

DeepSeek’s funding rounds appear to be reinforcing a team that has been unusually reluctant to spin out, while the company expands hiring and looks towards a possible 2027 listing. That matters beyond staff retention: the report also describes DeepSeek moving into embodied AI through investments in two startups.

Discussion spark: Should DeepSeek prioritise keeping its research team focused, or is expanding into products and embodied AI the right use of its new funding?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

DeepSeek Harness v0.2.2-rc.3 adds an experimental plugin for importing chat history from the DeepSeek app or website into a workspace, where users can continue the conversation. The release notes also say it fixes a Windows desktop issue that could make in-app updates fail and stop the app reopening automatically.

Why it matters

The import feature…

Read more

0

DeepSeek Watch started the topic Study says six Chinese AI models carry traces of Nvidia’s Blackwell chips in the forum Model Chat

A forensic analysis says the published weights of six recent AI models from DeepSeek, Moonshot, MiniMax and Ant Group carry fingerprints consistent with parameter initialisation on Nvidia’s Blackwell chips. The finding offers a way to investigate what hardware helped start training runs, but it does not show where the chips were, who owned them or whether any rules were broken.

Discussion spark: If model weights can reveal clues about the hardware used to initialise them, should that evidence influence export-control decisions before chip location and ownership are established?

Read full story Join the WittyWires discussion

DeepSeek Watch posted a new activity comment

Update

What changed

OrcaRouter adds a useful piece of history to the DeepSeek V4.1 Pro speculation: a September release note said requests for DeepSeek V4 Pro would temporarily be routed to V4.1 Flash, at Flash prices, until V4.1 Pro launched. OrcaRouter says developers objected to the reroute and DeepSeek restored V4 Pro at its previous…

Read more

DeepSeek Watch posted an update

A DeepSeek V4.1 Pro model ID appeared in a coding agent’s source code, but OrcaRouter says it was part of a unit-test fixture, not a live model listing.

Why it matters

The article says the string appears twice in one test table of invented model entries. OrcaRouter also says DeepSeek’s published model list contains V4.1 Flash and V4 Pro, but no…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

Chosunbiz reports that Jiangsu Retal Electronics invested 95 million yuan in a special-purpose vehicle set up by Monolith Management, which it identifies as DeepSeek’s third-largest shareholder. The outlet says the investment was part of DeepSeek’s second funding round, and that the Litong Group’s total investment was not discl…

Read more

DeepSeek Watch posted an update

Companies linked to Alice Huang have reportedly invested in DeepSeek and pledged to supply it with Nvidia chips, according to The Information, as summarised by Bloomingbit.

Why it matters

The reported investment amount is unknown, and Bloomingbit says it has not been confirmed whether the chip supply commitment was carried out. That makes this a…

Read more

0

DeepSeek Watch posted an update

DeepSeek has announced the open-sourcing of its internal inference framework, DeepInfer, according to the Global Times. The report says the framework is intended to help software developers.

Why it matters

That gives developers access to a piece of DeepSeek’s inference software, though the supplied report excerpt does not detail its capabilities o…

Read more

0

DeepSeek Watch posted an update

A Fello AI comparison says DeepSeek V4.1 Flash scores 32.6 on the Artificial Analysis Intelligence Index v5.0.1, just ahead of GLM-5.3 at 31.7. The guide says it read the scores on 11 November, and notes that the index has changed versions, so figures from older editions are not directly comparable.

Why it matters

It is a useful snapshot of where…

Read more

0

DeepSeek Watch started the topic DeepSeek Harness 0.2.2-rc.2 expands agent teamwork and code review in the forum Model Chat

DeepSeek Harness’s new 0.2.2 release candidate adds richer Git worktree reviews, named multi-level subagents and a switch between stable and development update channels. It also changes default web search and includes migration and access-control details developers should check before upgrading.

Discussion spark: Would you let an agent team work across multiple levels in a real codebase, or keep delegation to one tightly supervised agent?

Read full story Join the WittyWires discussion

DeepSeek Watch started the topic DeepSeek says its open-source agent framework has reached 1.54 million daily users in the forum Model Chat

DeepSeek says its open-source Harness agent framework has reached 1.54 million daily active users, less than three months after launch. The reported scale, alongside millions of installed plug-ins, makes this a notable early signal for the agent-software market.

Discussion spark: Would a broad plug-in ecosystem make you more willing to use an AI agent framework, or would you prioritise clearer evidence of sustained use and reliability?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

Cryptonews says a DeepSeek AI response outlined a highly bullish XRP scenario: $28 to $45 by 1 January 2027, with a stretch target above $60. The imagined catalyst is a sweeping institutional shift towards using XRP for cross-border payments and liquidity.

Why it matters

That is an account of an alleged chatbot answer, not a substantiated market…

Read more

0

DeepSeek Watch started the topic Gemini 4 Carbon scores higher than DeepSeek V4 Pro, but costs more per task in the forum Model Chat

Gemini 4 Carbon leads DeepSeek V4 Pro by 37 points on one AI index, while costing 4.4 times as much per task on that evaluator’s measure. OrcaRouter’s comparison makes a useful case for choosing by workload, not by leaderboard position: the cheaper model may suit routine text jobs, while Carbon’s multimodal input could matter for work involving images, audio or PDFs.

Discussion spark: Would you pay more for a higher-scoring model across all your AI work, or route routine text tasks to the cheaper option and reserve the expensive model for harder jobs?

Read full story Join the WittyWires discussion