# DeepSeek V4.1-Flash-1017 Scores 77.0 on DeepSWE and Opens a Paid Search API. Free Agent Calls End 31 October

The new checkpoint replaces the September Flash model under the same deepseek-flash name and price. The bigger change for agent builders is first-party web search at $0.75 per 1,000 calls, free inside three harnesses for two more weeks.

By Priya Nair (AI Industry Reporter) — published 2026-10-19, updated 2026-10-19
Source: https://aiscoutdaily.com/news/deepseek-v4-1-flash-1017-search-fetch-api-pricing

## TL;DR
DeepSeek released V4.1-Flash-1017 on 17 October 2026, an updated checkpoint now served on its API as deepseek-flash at an unchanged $0.30 per million cache-miss input tokens and $1.20 per million output tokens at peak, half that off-peak. DeepSeek reports 77.0 on DeepSWE v1.1, up from 74.2, at just under $0.20 per task against about $0.50 for the September model, but Claude Opus 5 still leads its own table on Terminal-Bench 4.0 (51.8 vs 42.3) and Humanity's Last Exam (56.3 vs 42.6). DeepSeek also opened Search and Fetch APIs at $0.75 per 1,000 successful calls each; calls made through DeepSeek Harness, Claude Code and Codex are free until billing starts on 31 October at 16:00 UTC.

Five weeks after V4.1-Flash, DeepSeek has shipped a new post-training checkpoint of the same model. There is no new model name to switch to and no price change. If you already call deepseek-flash, you are on the new checkpoint now, which is the main reason to re-run your own tests this week.

## What changed on 17 October

- Model: DeepSeek-V4.1-Flash-1017 is live on DeepSeek's website, app and API. On the API it is served as deepseek-flash, with a 1M-token context window and up to 384K output tokens.
- Legacy names: deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but those models are retired and requests go to the 1017 checkpoint at the Flash price.
- Weights: published on Hugging Face and ModelScope. The Hugging Face model card lists an MIT licence.
- Claims: DeepSeek calls it a Pareto improvement, with higher scores, faster task completion and shorter outputs than the September release. It says the model was trained with reinforcement learning using self-feedback for end-to-end project work, and it recommends DeepSeek Harness v0.2.1-rc.1 or later.

## The benchmarks, as DeepSeek reports them

| Benchmark | V4.1-Flash-1017 | V4.1-Flash (Sep) | Claude Opus 5 | GPT 5.6 Sol |
| --- | --- | --- | --- | --- |
| DeepSWE v1.1 | 77.0 | 74.2 | 74.0 | 73.0 |
| Terminal-Bench 4.0 | 42.3 | 31.2 | 51.8 | 39.9 |
| Automation-Bench v1.0.6 | 60.5 | 54.8 | 50.3 | 45.8 |
| Agents' Last Exam | 35.4 | 31.8 | 28.6 | 26.7 |
| ProgramBench | 29.1 | 20.3 | 37.0 | 23.0 |
| NL2Repo-Bench | 68.2 | 65.4 | 75.3 | 56.8 |
| Humanity's Last Exam (no tools) | 42.6 | 36.8 | 56.3 | 44.5 |

*DeepSeek's own figures, as tabulated by MarkTechPost. Terminal-Bench 4.0 was run in Claude Code and DeepSWE in Mini-SWE-V2. Not independently verified.*

MarkTechPost counts gains over the September model on all 11 benchmarks DeepSeek reported, by 2.0 to 11.1 points. The pattern from the earlier release still holds: the model leads on agentic coding and automation tests and trails Claude Opus 5 by a wide margin on Terminal-Bench, repository-scale work and the knowledge-heavy Humanity's Last Exam.

## The cost per task is the real headline

Token prices did not move, but DeepSeek says the new checkpoint writes shorter outputs. The Decoder reports DeepSeek's DeepSWE cost chart: 77.0% at just under $0.20 per task, and above 72% for about $0.10 on the low reasoning setting. The September model scored 74.2% at around $0.50 per run. Those are DeepSeek's measurements on its own chosen harness, so treat them as a direction to test rather than a number to budget on.

| deepseek-flash, per 1M tokens | Cache-hit input | Cache-miss input | Output |
| --- | --- | --- | --- |
| Off-peak | $0.003 | $0.15 | $0.60 |
| Peak | $0.006 | $0.30 | $1.20 |

*DeepSeek pricing page, checked 19 October 2026. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays.*

## Search and Fetch: what they cost and when the free window ends

- What they do: Search returns ranked web results for a query, and Fetch returns a single page as Markdown. DeepSeek says its in-house engine serves hundreds of millions of requests a day.
- Price: $0.75 per 1,000 successful requests for each endpoint. MarkTechPost reports that empty results and failures are not billed, and that the limit is 5 queries per second per account for each API.
- Free window: server-side Search and Fetch calls made through DeepSeek Harness, Claude Code or Codex are currently free for DeepSeek API users. The release note says billing starts on 31 October 2026 at 16:00 UTC; the pricing page gives the same moment as 00:00 Beijing time on 1 November.

## Two details most coverage got wrong or missed

- Licence: MarkTechPost's summary says the 1017 weights' licence was not disclosed. The Hugging Face model card for DeepSeek-V4.1-Flash-1017 lists MIT.
- V4-Pro is back on the price list. DeepSeek retired V4-Pro on 14 September and routed it to Flash. Its pricing page now lists deepseek-v4-pro as a separate model, DeepSeek-V4-Pro-0813, at $0.66 per million cache-miss input tokens and $1.98 per million output tokens off-peak ($1.32 and $3.96 at peak), with a concurrency limit of 500 against 2,500 for Flash and no vision support. DeepSeek has not published a note explaining the change.

## What this means if you build on DeepSeek

- Your deepseek-flash traffic already moved to 1017. Re-run your own evaluations, and check output length, since shorter answers are part of the pitch.
- If your agents use Claude Code or Codex against DeepSeek's API, web tool calls start costing money on 31 October. At $0.75 per 1,000, a busy agent fleet should check its call volume before then.
- If you need repository-scale changes or broad knowledge, DeepSeek's own table still shows a clear gap to Claude Opus 5. The price makes Flash worth testing, not an automatic swap.

## Sources
- [DeepSeek API Docs: DeepSeek-V4.1-Flash-1017 release (Oct 17, 2026)](https://api-docs.deepseek.com/news/news261017/)
- [DeepSeek API Docs: Models & Pricing](https://api-docs.deepseek.com/quick_start/pricing)
- [Hugging Face: deepseek-ai/DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017)
- [MarkTechPost: DeepSeek Just Released V4.1-Flash-1017, Scores 77.0 on DeepSWE, Opens Search API (Oct 17, 2026)](https://www.marktechpost.com/2026/10/17/deepseek-just-released-v4-1-flash-1017-scores-77-0-on-deepswe-opens-search-api/)
- [The Decoder: DeepSeek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs](https://the-decoder.com/deepseek-v4-1-flash-1017-beats-larger-rivals-in-agentic-tasks-while-once-again-massively-cutting-per-task-costs/)