DeepSeek V4.1-Flash + Search/Fetch API: What Changed Oct 17, and the Search API Free Window Until Nov 1

The short version: on October 17, 2026, DeepSeek shipped V4.1-Flash-1017 — a faster, sharper version of its cheap flagship model — and two brand-new API endpoints, Search and Fetch, that give any app live web access. Token prices didn’t move (still $0.15/$0.60 per 1M tokens off-peak, $0.30/$1.20 peak — per official pricing). And there was a deadline that mattered: Search/Fetch calls made through DeepSeek Harness, Claude Code, or Codex were free until November 1, 2026 (midnight Beijing time) — that window has now closed, and those calls are billed at $0.75 per 1,000 successful requests. The model itself has always been billed per token — only the Search/Fetch API calls were free in that window.

Update (Nov 1, 2026): the free window has ended — billing began at 9:00 AM Pacific on October 31 (midnight November 1, Beijing time). Server-side Search/Fetch calls through Harness, Claude Code, or Codex are now billed at $0.75 per 1,000 successful requests. The “free window” language elsewhere in this article describes the window before it closed.

What actually changed

The model (V4.1-Flash-1017). Per DeepSeek’s official docs: benchmark scores up across the board, outputs shorter (less rambling = lower bills), token price unchanged — still $0.15/$0.60 per 1M input/output tokens off-peak, $0.30/$1.20 at peak hours (per official pricing, multiple sources). Correction (Oct 22): the “free until Nov 1” window applies only to the Search/Fetch APIs (below), not to the model itself — the model has always been billed per token. The 552B-parameter open weights went up on Hugging Face the same day under the MIT license — meaning anyone can download and run it.

The two new APIs. Think of them as “Google in a box” for your app:

• Search API — send a query, get web results back.
• Fetch API — send a URL, get the page’s content back.

Pricing: $0.75 per 1,000 successful requests. Read that unit carefully — it’s per request, not per token. And “successful” does real work here: only HTTP 200 responses that return actual results are billed. Failed, empty, or timed-out calls cost nothing. What that means for you: a research agent doing 1,000 searches costs $0.75, predictable to the cent.

The Nov 1 free window — now closed (and how billing works now)

DeepSeek’s official docs say: when you use the DeepSeek API through DeepSeek Harness, Claude Code, or Codex, the server-side Search/Fetch calls those tools make were free until 00:00 Beijing time on November 1, 2026; per-request billing applies since then. Call the endpoints directly and you’re billed from the first request. (Updated Oct 23, 2026: price unchanged at $0.75 per 1,000 successful requests; account-level rate limits have been raised to Search 100 QPS / Fetch 25 QPS per the current official docs — generous enough that a monitoring bot hitting them every few seconds stays well under the cap.)

Mini-tutorial — try the Search API in 3 steps:

1. Get a DeepSeek API key at platform.deepseek.com (the chat app itself is free; the key is for API use).
2. Send a search request (replace YOUR-KEY):

curl https://api.deepseek.com/v1/search \
  -H "Authorization: Bearer YOUR-KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "latest NVIDIA GPU prices October 2026", "count": 5}'

3. You’ll get back structured results (title, URL, snippet per hit). Pipe them into your app, your agent, or a spreadsheet.

(Endpoint shape follows DeepSeek’s documented pattern — confirm the exact path in the official Search guide before wiring it into production.)

The money angle

Sell “live-data agents” — the free window has closed, but the math still works. Plenty of small businesses want an agent that watches the web — competitor price tracking, new-listing alerts, daily industry briefings. Those agents need search, and search now has a clean, cheap, official API.

The play: build the agent on DeepSeek Harness. The Search calls it makes are now billed, but at $0.75 per 1,000 successful requests your demo and first client pilots cost pennies. Charge $200–$600/month per monitoring client. Your cost per client is still tiny (1,000 searches/day = ~$22/month) — the margin survives; the free window only made the validation phase free.

7-day validation: Day 1–2: get a key, run the curl above, wire results into a Google Sheet. Day 3: pick one niche (e.g., “new Airbnb listings in Scottsdale”) and build a daily-alert demo. Day 4–5: show it to 10 people in that niche. Day 6–7: one “I’d pay for this” = validated.

References

• DeepSeek Search API guide (official): https://api-docs.deepseek.com/guides/search/
• DeepSeek platform (official): https://platform.deepseek.com