DeepSeek V4.1-Flash-1017 Update Adds Open Weights and Search API
DeepSeek has released V4.1-Flash-1017, an MIT-licensed update to its Flash model with better coding scores, plus a new paid Search and Fetch API for developers.

In this article
DeepSeek released DeepSeek V4.1-Flash-1017 on October 17, an updated checkpoint of its V4.1-Flash model with open weights under the MIT license. Alongside the model, the Chinese AI lab launched a paid Search and Fetch API that lets developers give their apps live web access. According to DeepSeek's release notes, the update is already live on its website, app and API.
What's new in DeepSeek V4.1-Flash-1017
DeepSeek describes the release as delivering "stronger benchmark performance, faster task completion, and shorter outputs than the previous version." Its Hugging Face model card calls it a minor update to V4.1-Flash, but the published scores show sizable jumps on several agent and coding tests.
Key details from the model card and release notes:
- License: MIT for both the code repository and the model weights, so businesses can use and modify it with few restrictions.
- Size: the model card lists 763 billion total parameters, with weights published in 8-bit formats.
- Context: a recommended context window of 1 million tokens and a suggested maximum output of 256K tokens or more.
- Inputs: text and images (image-text-to-text).
- Training focus: reinforcement learning with self-feedback aimed at end-to-end projects such as building games and 3D models. DeepSeek's demos include a 3D model of the Forbidden City and a Portal 2 recreation in the Godot engine.
- Tooling: optimized for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, with inference support listed for vLLM and SGLang.
Benchmark changes vs. the September release
DeepSeek published side-by-side scores against the original V4.1-Flash. A selection:
| Benchmark | V4.1-Flash-1017 | V4.1-Flash |
|---|---|---|
| DeepSWE v1.1 | 77.0 | 74.2 |
| Terminal-Bench 4.0 | 42.3 | 31.2 |
| HLE (no tools) | 42.6 | 36.8 |
| ProgramBench | 29.1 | 20.3 |
| Apex Bench v1.1 | 72.1 | 64.6 |
The biggest gain is on Terminal-Bench, which measures how well a model completes tasks in a command-line environment, a core skill for coding agents. As always, these are vendor-reported figures.
The new Search and Fetch API
The second half of the announcement may matter more to developers. DeepSeek's Search and Fetch API runs on the company's in-house search engine and costs $0.75 per 1,000 calls per endpoint. Search and Fetch calls made through server-side tools in DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until billing begins on October 31, 2026, at 16:00 UTC, according to the release notes.
Web access is one of the building blocks of AI agents, which need fresh information to research, compare and act. Bundling a search API with an open-weight model makes DeepSeek a more complete option for teams building those tools.
Why it matters
DeepSeek continues to ship competitive models under one of the most permissive licenses available, which keeps pressure on closed-model providers on both price and openness. A 763-billion-parameter model is not something you will run on a laptop, but MIT licensing means cloud providers and companies with their own GPU clusters can host, fine-tune and redistribute it.
For individual users, the practical change is simpler: the DeepSeek app and website now run the updated model. Whatever assistant you use, clear instructions still make the biggest difference, so our guide on how to write better AI prompts applies here too.
Bottom line
V4.1-Flash-1017 is an incremental version number with meaningful gains for coding and agent tasks, and the new Search and Fetch API gives developers a cheap way to add live web results. If you build with open models, it is worth testing before the free search period ends on October 31.
Source: DeepSeek API Docs
Add The Tech Wonder as a preferred source on Google



0 Comments
Join the discussion. Comments are moderated and your email is never published.