19 October 2026
DeepSeek prices its search and fetch tool at $0.75 per 1,000 requests
Today was mostly about tools and rules. DeepSeek put a price on web search for its models, and a free WhatsApp connector appeared. Anthropic was in the news for a safety bill, outside oversight and a share pledge, and Meta changed its terms on web scraping.
Tools and products
DeepSeek search and fetch costs $0.75 per 1000 requests (opens the source in a new tab)
DeepSeek's API docs list a price of $0.75 per 1000 successful requests for its search and fetch feature.
Why it matters: If you build agents that browse the web, work out your monthly request volume and compare this price with what you pay now.
api-docs.deepseek.com via Hacker News · 18 Oct
WhatsApp MCP now runs free on Cloudflare (opens the source in a new tab)
A new open project on GitHub lets you run a WhatsApp MCP connector on Cloudflare at no cost. It was shared on Hacker News.
Why it matters: If your clients live on WhatsApp, this could let an AI assistant work with those chats, but test it carefully before it touches customer messages.
github.com via Hacker News · 18 Oct
Sphinx quizzes you on AI-written code before shipping (opens the source in a new tab)
Sphinx is a hook that asks you questions about the code your AI wrote before you ship it. It was shared on Hacker News.
Why it matters: If you or your developer ship AI-written code, a check like this helps make sure someone understands it before clients rely on it.
sphinx-six-theta.vercel.app via Hacker News · 18 Oct
Security and safety
Meta removes language that raised scraping and AI training concerns (opens the source in a new tab)
Meta removed wording from its terms that had raised concerns about scraping and AI training. A Meta spokesperson said the update did not change how Meta crawls the web.
Why it matters: If you publish content online, check how your content may be used for AI training and keep an eye on terms that change.
TechCrunch · 17 Oct
Anthropic's proposed AI safety bill has high trigger levels (opens the source in a new tab)
Tom's Hardware reports that Anthropic's proposed AI safety bill only applies after 50 deaths or $1B in damages.
Why it matters: Rules for AI are still being shaped, so keep track of them if you use AI with client data or in regulated work.
tomshardware.com via Hacker News · 18 Oct
Anthropic gives outside AI watchdog employee-like access
Enterprise
Disney agrees to hire most of Character.AI's workforce (opens the source in a new tab)
A filing says the vast majority of Character.AI employees offered jobs at Disney accepted. It is a stalking horse bid, so another company could still offer more.
Why it matters: If you use a smaller AI product, remember that deals like this can change who runs it, so have a backup option.
TechCrunch · 18 Oct
Anthropic commits 5% of its shares to a public benefit fund (opens the source in a new tab)
Anthropic said it will commit 5% of its shares to a public benefit fund, according to TechCrunch.
Why it matters: This does not change your Claude plan today, but it shows how one of your AI vendors is setting itself up as a company.
TechCrunch · 18 Oct
Anthropic writes about how employment is changing at the company (opens the source in a new tab)
Anthropic published a post on how employment is changing at Anthropic. It drew a lot of discussion on Hacker News.
Why it matters: It is worth a read if you are deciding which tasks in your own team to hand to AI.
Also worth knowing
DeepSeek V4.1 Flash 1017 (Max) listed on Artificial Analysis (opens the source in a new tab)
The independent site Artificial Analysis has a page for a DeepSeek model called V4.1 Flash 1017 (Max). It was shared on Hacker News.
Why it matters: If you use DeepSeek models, check the page for how the model scores before you switch.
artificialanalysis.ai via Hacker News · 19 Oct
How far will AI assistants go for a refund? (opens the source in a new tab)
Vals.ai tested how AI assistants behave when asked to get a refund they cannot get.
Why it matters: If you let an AI assistant talk to customers or suppliers for you, test how it acts when the answer is no.
vals.ai via Hacker News · 18 Oct
Hybrid Reasoning Agents think in latent space and act in text (opens the source in a new tab)
A new paper describes LLM agents that think in latent space and act through text.
Why it matters: This is early research, but it points to how future agents may work, so just keep an eye on it.
arxiv.org via Hacker News · 18 Oct