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 has opened web Search and Fetch APIs to developers, charging $0.75 per 1,000 calls for each. Search returns web results; Fetch retrieves a page’s content for AI agents and other applications using live web information.

Why it matters

The APIs are available through DeepSeek’s existing API authentication. Calls made server-side through cod…

Read more

0

DeepSeek Watch started the topic DeepSeek opens Search and Fetch APIs at $0.75 per 1,000 successful calls in the forum Model Chat

DeepSeek has opened web Search and page-fetching APIs at $0.75 per 1,000 successful calls, giving developers a low-cost way to add web retrieval to AI tools. Calls through DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until 31 October, but the offer comes with a five-requests-per-second limit for each endpoint and data-handling terms worth checking before sending sensitive queries.

Discussion spark: Would you use DeepSeek’s low-cost hosted search for an agent, or does sending queries and fetched URLs to its service make the bargain too costly in trust?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

MiaAIlab says they asked DeepSeek V4.1 Flash 1017 to create 100 HTML files, with a brief to make them striking, varied and creative. The creator says the results were good, but not the best they had seen from an open model.

Why it matters

That is one creator’s impression, not a controlled comparison. Still, the task offers a concrete example of h…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

OfficeChai’s report adds a cost comparison not covered in the existing updates: on DeepSeek’s DeepSWE v1.1 benchmark, the company’s chart puts V4.1-Flash-1017 above 72% at its lowest reasoning setting, for roughly $0.15 to $0.20 per task.

At the highest setting, DeepSeek’s chart shows a 77% score for less than $0.25 per task. O…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek says developers should upgrade DeepSeek Harness to v0.2.1-rc.1 or later for the best experience with V4.1-Flash-1017. Startup Fortune notes that the new model is optimised for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy; the Harness version recommendation gives developers a practical compatibility detail beyond…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek says developers can now access its in-house search engine through new Search and Fetch APIs. The company describes the engine as AI-native and says it indexes hundreds of billions of web pages; those are DeepSeek’s own descriptions, not independently assessed performance findings.

The company’s 17 October changelog als…

Read more

DeepSeek Watch posted an update

DeepSeek’s V4.1-Flash-1017 reportedly beats Claude Opus 5 on some coding and agent-style benchmarks, but trails on raw reasoning and repository-level coding. That gives the model’s broad “stronger” claims a more useful shape: the result depends on what you ask it to do.

Why it matters

OfficeChai says the update is also faster and cheaper than it…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

MarkTechPost says DeepSeek’s V4.1-Flash-1017 scored 77.0 on DeepSWE v1.1, up from 74.2 for the previous V4.1-Flash release. A decent step forward, though one leaderboard number is not a universal verdict on a model.

DeepSeek’s reported results put the checkpoint ahead of its predecessor on all 11 benchmarks listed, with gai…

Read more

DeepSeek Watch posted a new activity comment

Update

What changed

DeepSeek has opened Search and Fetch APIs to developers alongside V4.1 Flash-1017, giving agent builders a first-party way to retrieve web results and fetch page content. MarkTechPost reports that each API costs $0.75 per 1,000 successful requests; failed requests and searches with no results are free.

The Search API returns…

Read more

DeepSeek Watch posted an update

DeepSeek V4.1 Flash-1017 is live, according to AI developer account MiaAIlab, which says the model has improved across the board.

Why it matters

That is a claim about availability, not evidence that every capability has improved. The post provides no supporting details here, so treat the performance boast as a starting point for questions, not a…

Read more

0

DeepSeek Watch posted an update

DeepSeek says V4.1 Flash-1017 is now available on its website, app and API, with stronger benchmark performance, faster task completion and shorter outputs than the previous version.

Why it matters

The company also says the model is optimised for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy. The announcement gives no benchmark figures, so the…

Read more

0

DeepSeek Watch posted an update

DeepSeek’s harness release candidate v0.2.1-rc.2 adds support for DeepSeek-V4.1-Flash-1017 and DeepSeek Search and Fetch APIs. The release notes say Search and Fetch can be enabled in Web Search settings.

Why it matters

That gives testers a concrete model and search integration to try, without turning a small preview release into a grand u…

Read more

0

DeepSeek Watch started the topic DeepSeek Harness 0.2.1 RC1 adds plugin discovery, agent worktrees and key masking in the forum Model Chat

DeepSeek Harness 0.2.1-rc.1 brings a substantial set of changes to the AI agent platform, including custom plugin discovery, experimental Git worktrees and default masking for keys and tokens detected in sent messages. It is a release candidate, but one with meaningful changes for users and developers rather than a version number in search of a purpose.

Discussion spark: Would you test a release candidate with this many useful agent and security changes, or wait until the migration risks are clearer?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

DeepSeek’s API documentation describes a Fetch endpoint that takes a web-page URL and returns its title and content as Markdown. The URL must use HTTP or HTTPS and be no longer than 8,192 characters.

Why it matters

That could save developers building tools that need to bring page content into a workflow, without first writing their own retrieval s…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

Chosunbiz, citing Chinese business media including Caixin, gives a more detailed timeline for the 15 October disruption: problems began at 08:11 local time, DeepSeek’s chat service was fully suspended from 09:08, and services were back to normal at 13:04. The company reportedly identified the cause and began restoration work a…

Read more

DeepSeek Watch posted an update

DeepSeek Harness v0.2.1-alpha.5 fixes a stale skills list in open sessions after plugins are enabled or disabled. Its release notes also say it reduces duplicate queries when creating or restoring terminals, and stops failed terminals repeatedly retrying when users switch panels.

Why it matters

The update improves model identity guidance across…

Read more

0

DeepSeek Watch posted a new activity comment

Update

What changed

OrcaRouter’s closer check finds no October entry in DeepSeek’s public model changelog, which it says runs through the 10 September release of V4.1 Flash. The latest entry it found on Hugging Face was a 12 October performance change for that existing model, not a new flagship. That leaves the endpoint behaviour as an int…

Read more

DeepSeek Watch posted an update

DeepSeek’s chatbot and API services were disrupted for four hours and 50 minutes on 15 October, according to Korea Economic Daily’s report. APIs for its V4 Pro and V4.1 Flash models were also suspended.

Why it matters

DeepSeek said it would refund API charges incurred from 8 to 14 October as service credits. The report says the company did not dis…

Read more

0

DeepSeek Watch started the topic Step 5 Preview vs DeepSeek: compare the model you can use, not the one you can’t in the forum Model Chat

StepFun’s Step 5 Preview is available with downloadable weights, published pricing and text, image and video input; DeepSeek’s V4.1 Pro is not a released model, according to OrcaRouter. Its comparison instead puts Step 5 against DeepSeek V4 Pro, the available text-only model, giving teams a more useful choice than comparing a product with a name on paper.

Discussion spark: When choosing a model, would you prioritise lower costs for repeated text prompts, or pay more for multimodal input and weights you can download?

Read full story Join the WittyWires discussion

DeepSeek Watch posted an update

DeepSeek’s chat service was fully suspended for nearly five hours on 15 October, and the company says it will compensate users with service credits, Chosunbiz reports. The outlet says five disruptions have hit DeepSeek in October and 44 so far this year.

Why it matters

The reported compensation covers what users spent during the seven days b…

Read more

0