MOFCOM and MOST Announcement No. 48 would add 40 technologies in AI, biomedicine and solar to China's export catalogue. Exporters must already register listed technologies before export while the draft is open for comment.
DeepSeek stops charging for model-triggered search and fetch calls
EditorialBy TrustList Editorial
From 12:00 Beijing time on 18 November 2026, DeepSeek no longer bills the server-side search and web fetch calls its models trigger; their results are billed as model tokens, and the standalone Search and Fetch APIs stay at US$0.75 per 1,000 requests.
About DeepSeek stops charging for model-triggered search and fetch calls
DeepSeek stops charging for model-triggered search and fetch calls
18 November 2026: DeepSeek has changed how it bills web tools in its API. From 12:00 Beijing time (04:00 UTC) on the 18th, the server-side search and web fetch calls that a model triggers while answering a request are no longer charged as separate calls. The pages and results they return are added to the model's context and billed as ordinary token usage. Teams that run research or browsing agents on DeepSeek models should see a different shape to their bills from this week.
Not yet independently verified. The change, its start time and the prices come from DeepSeek's API pricing page, read on 19 November; the page carries no date of its own and the change was not yet in DeepSeek's change log. DeepSeek does not state what model-triggered calls cost before the change, so no saving is given here. We will update this when it can be confirmed, and remove this note.
Web tool calls now billed as tokens
The pricing page says the change covers the web tools in DeepSeek's own API, the web_search tool in its Anthropic-format API and the web_search tool in its Responses-style API. In each case, a search or fetch that the model decides to make is now free as a call, but the text it brings back counts as input tokens at the model's normal rate.
The standalone APIs, which developers call directly rather than through a model, keep their prices: US$0.75 per 1,000 successful requests for the Search API and the same for the Fetch API. File parsing and OCR cost US$0.225 per 1,000 pages. Text to speech is US$0.45 per 10,000 characters, currently halved to US$0.225 for a limited time, and speech recognition US$0.116 per hour, currently US$0.058.
Token prices and peak hours
The same page lists the two main models with off-peak and peak prices per million tokens. deepseek-flash, with a one million token context and up to 384,000 output tokens, costs US$0.15 off-peak and US$0.30 at peak for uncached input, US$0.003 and US$0.006 for cached input, and US$0.60 and US$1.20 for output. deepseek-v4-pro costs US$0.66 and US$1.32 for uncached input, US$0.022 and US$0.044 for cached input, and US$1.98 and US$3.96 for output.
Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour, weekends and those holidays are billed at the off-peak rate.
The change follows the release of DeepSeek's suite of application APIs, logged on 13 November, which made web search, file input, URL reading, model safety, speech recognition and text to speech available as separate services. The Search API itself opened on 17 October.
What teams using the API should do
Because search results are now billed as tokens, the cost of an agent depends on how much text each search pulls into the context, not on how many searches it makes. Check whether your client sets limits on the number or size of results, and watch input token use in the first days after the change. For workloads that only need search results and no model answer, the standalone Search API at a fixed price per request may still be cheaper.
Teams using DeepSeek through a cloud marketplace or reseller should check whether that provider has adopted the same billing. Where data residency matters, note that web tool calls run on DeepSeek's servers, so the queries and fetched pages pass through its infrastructure.
Sources
- DeepSeek API Docs: Models and Pricing, 18 November 2026
- DeepSeek API Docs: Change log, 13 November 2026
- Pandaily: DeepSeek removes separate model-triggered search and fetch fees, 19 November 2026
Categories & features
TrustList Weekly
The week in software and IT, in one email
The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.
You may be interested in
The open-weight model's terms grant rights only outside those three markets, and products above $50 million in annual revenue need Baidu's written authorisation first.
The attempt came in fewer than one call in 100,000 and was stopped by a sandbox with no route out. Cohere has published the traces; Alibaba has not responded publicly.
From 31 October new sign-ups lose access to kimi-k2.6 and kimi-k2.7-code; existing API users keep them until a retirement Moonshot plans for the end of December, and K3 bills cache writes separately.
Agent tool calls through DeepSeek Harness, Claude Code or Codex already use the new Search and Fetch endpoints for free; per-request billing starts at midnight Beijing time on 1 November.
The draft adds 40 technologies in AI, biomedicine and solar to the export catalogue. Registration applies at once, and comments close on 15 November.
San Francisco generative media platform each::labs says it raised a US$7 million seed round co-led by Revo Capital and the EBRD on 18 November 2026, after its annualised revenue run rate passed US$20 million in October.
A consultation paper under Act 709 would have controllers check vendors on training use, sub-processors and transfers abroad. It is a draft with no start date, and comments closed on 23 October.
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- More ChinaThe ranking for this subject
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.