✳ the wire · analysis

DeepSeek upgrades V4.1 Flash to the 1017 build: same price, higher scores on its own benchmarks

DeepSeek V4.1 Flashconfirmedby ArtificialWatch

Get launch alerts like this, free →

ArtificialWatch title card · verified against a primary source

DeepSeek upgraded V4.1 Flash to a new checkpoint, DeepSeek-V4.1-Flash-1017, on October 17. DeepSeek's changelog says the API id and the price do not change: deepseek-flash now serves the new model, and DeepSeek's web and mobile apps have switched to it too. The retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp also route to it now.

OPEN WEIGHTS

The weights are public on Hugging Face as deepseek-ai/DeepSeek-V4.1-Flash-1017 under the MIT license, the same license as the September 10 release. The repo's model config file is identical to the earlier one, so the architecture and the 763B size are unchanged: the new checkpoint changes the weights, not the shape of the model.

THE PRICE

Unchanged, says DeepSeek. Its pricing page charges $0.15 per million input tokens and $0.60 per million output off-peak, with cached input at $0.003. Peak hours cost double, $0.30 and $1.20: 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. OpenRouter lists the new build as deepseek/deepseek-v4.1-flash-1017 at $0.15 / $0.60. The context window is 1M tokens, with up to 384K output tokens.

WHAT CHANGED, BY DEEPSEEK'S OWN FIGURES

DeepSeek's changelog compares the new build with the September 10 release:

Terminal-Bench 4.0: 42.3, up from 31.2 ProgramBench: 29.1, up from 20.3 Humanity's Last Exam without tools: 42.6, up from 36.8 DeepSWE v1.1: 77.0, up from 74.2 NL2Repo-Bench: 68.2, up from 65.4

It also lists new scores on Agents' Last Exam (35.4), Apex Bench v1.1 (72.1) and BabyVision with tools (92.5). DeepSeek says the new model finishes tasks faster and with shorter output, so the same job costs less. These are DeepSeek's numbers; no third-party board has scored the 1017 build yet.

ALSO TODAY

The same changelog opens DeepSeek's in-house search engine to developers as an API, the one that already powers search in its own app and API.

THE MARKET

Polymarket's "Next DeepSeek Flash released by...?" market counts later dated builds of V4.1 Flash, so this update settles it. It was trading at 99.95% Yes for October 31 when we checked. Our own hunt for the next Flash stays on a V4.2 or later.

WHAT WE DO NOT KNOW

How independent boards will score the 1017 build, and whether they will list it apart from the September 10 release.

Source: DeepSeek API changelog (Oct 17 entry) and pricing page, read directly · Hugging Face (deepseek-ai/DeepSeek-V4.1-Flash-1017) · OpenRouter catalog · Polymarket · @LuminaBench, @Mr_Salio and @scaling01 on X ↗ · DeepSeek V4.1 Flash tracker · the bench index

sweeping every 60 seconds

Hear the minute DeepSeek’s next model drops — not the minute we write it up.

Claude Opus 5 went live at 16:51 UTC. The alert was in subscribers' inboxes at 16:52.

  • Free forever
  • No card
  • Unsubscribe in one click

← back to the wire