In this article
  1. What happened
  2. Why it matters
  3. The prices, per 1M tokens (deepseek-flash)
  4. What to do

TL;DR: On Oct 17, 2026, DeepSeek moved its deepseek-flash API model to a new build, DeepSeek-V4.1-Flash-1017. The calling method and prices are unchanged, so apps that use deepseek-flash (or the retired deepseek-v4-flash names) switched to the new model without a code change. Rerun your own tests. Separately, Search and Fetch calls that DeepSeek API users make through server-side tools in Claude Code, Codex and DeepSeek Harness are free only until Nov 1, 2026, 12:00 a.m. Manila time.

What happened

DeepSeek’s API change log, dated Oct 17, 2026, says: “DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017. The official web and APP models have also been updated to the new version. The API calling method and pricing remain unchanged — simply use deepseek-flash to access the latest model.”

The API docs add that the legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp “are still accepted, but the corresponding models have been retired; their requests are served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price.”

DeepSeek also published the model weights on Hugging Face under the MIT License. Hugging Face lists the model at 763B parameters.

In DeepSeek’s model card, the new build scores higher than the previous V4.1-Flash on all 11 benchmarks listed, and DeepSeek says it reduces task run time and output length. Two examples from its model card: DeepSWE v1.1 rose from 74.2 to 77.0, and Terminal-Bench 4.0 from 31.2 to 42.3. These are DeepSeek’s own results. It ran Terminal-Bench 4.0 in the Claude Code harness and DeepSWE in the Mini-SWE-V2 harness.

On the same day, DeepSeek opened its Search and Fetch APIs to all developers at $0.75 per 1,000 successful requests for each endpoint.

Why it matters

The model behind deepseek-flash changed on DeepSeek’s side. If your product, agent or automation calls that name, its answers, output length and tool-calling behavior may differ from last week, even though you didn’t deploy anything. Shorter outputs can lower your bill, but they can also break a parser or prompt that expects a certain format.

DeepSeek’s Models & Pricing page lists two model names, deepseek-flash and deepseek-v4-pro, and doesn’t list a separate name for the previous Flash build.

The prices, per 1M tokens (deepseek-flash)

Off-peak Peak
Input, cache hit $0.003 $0.006
Input, cache miss $0.15 $0.30
Output $0.60 $1.20

Source: DeepSeek, Models & Pricing (checked Oct 18, 2026). Context length is 1M tokens, maximum output is 384K tokens, and the listed concurrency limit is 2,500.

Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, excluding Chinese public holidays. In Manila time, that’s 9 a.m. to 12 p.m. and 2 p.m. to 6 p.m. on weekdays. All other hours are off-peak, including weekends and Chinese public holidays, and off-peak rates are half the peak rates.

What to do

  1. Rerun your evals. If you call deepseek-flash, deepseek-v4-flash or deepseek-v4-flash-vision-exp, you’re already on the 1017 build. Run your usual test prompts, and check output format, length and tool calls before trusting earlier results.
  2. Move batch work off-peak. Jobs that can wait, such as nightly summaries, bulk tagging or report generation, cost half as much outside the peak windows above.
  3. Budget for Search and Fetch. If you use DeepSeek as the backend in Claude Code, Codex or DeepSeek Harness, DeepSeek says their server-side search and fetch calls are currently free. Billing starts Oct 31, 2026, at 16:00 UTC (Nov 1, 12:00 a.m. Manila time), at $0.75 per 1,000 successful requests per endpoint. Check how often your agents search before then.
  4. Update DeepSeek Harness. DeepSeek recommends version 0.2.1-rc.1 or later for the new model.
  5. Treat your API key like any provider key. DeepSeek deducts usage from your topped-up or granted balance and recommends topping up based on actual usage. If you paste a DeepSeek key into a third-party coding tool, rotate it if that tool has a security incident, as covered in our OpenCode breach story.

If you’re weighing a pay-per-token API against a flat coding subscription, our Claude and Codex usage limits guide explains how those plans meter usage.

Sources & method

Sources

Method. Last checked October 18, 2026. Quotes, prices, limits and dates come from DeepSeek’s own docs and model card. Benchmark scores are DeepSeek’s and were not reproduced by us. Manila times are converted from the UTC times DeepSeek publishes. No affiliate links on this page. How we work.

Spotted an error? See our corrections policy and tell us.