DeepSeek updates V4.1 Flash and opens Search and Fetch APIs

DeepSeek has released V4.1-Flash-1017, an update to its V4.1 Flash model that improves coding and AI agent performance. According to the company, the new version completes tasks faster and generates fewer output tokens, reducing the cost of completing tasks.The model weights are available on Hugging Face under the MIT license. The update is also available through DeepSeek's website, app, and API.

Benchmark performance

In DeepSeek's evaluations, V4.1-Flash-1017 scored 77.0 on DeepSWE v1.1, compared with 74.2 for the previous version. It also surpassed Claude Opus 5 (74.0) and GPT-5.6 Sol (73.0) on this benchmark.The updated model improved across several other benchmarks, although Claude Opus 5 continues to lead on Terminal-Bench 4.0, ProgramBench, NL2Repo-Bench, and Humanity's Last Exam (HLE).

Source: DeepSeek's official model evaluations. Results may depend on the evaluation framework and configuration.These results show improvements over the previous V4.1 Flash across all five benchmarks. The largest gain among those listed is on Terminal-Bench 4.0, where the score increased from 31.2 to 42.3.DeepSeek also reports shorter task execution times and reduced output length. The model's API pricing remains unchanged, so the potential savings come from completing tasks with fewer generated tokens.

DeepSeek opens Search and Fetch APIs to developers

Alongside the model update, DeepSeek has made its Search API and Fetch API available to developers. Both are powered by the company's own search engine.The two APIs serve different purposes:

Search API accepts a search query and returns structured results containing relevant web pages.
Fetch API retrieves the content of a specified web page and returns it as Markdown.

Developers can use these tools to give AI agents access to current information, retrieve source material, and build retrieval-augmented generation (RAG) systems.

API pricing

Both APIs cost $0.75 per 1,000 successful requests.

Only requests that successfully return results are billed. Empty search results, failed requests, and timeouts are not charged.At launch, Search and Fetch calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex were offered free of charge to DeepSeek API users.DeepSeek scheduled per-request billing for these calls to begin on November 1, 2026, at 00:00 Beijing time. Developers should check their current billing status before relying on free tool access.Full endpoint specifications and usage examples are available in the official Search and Fetch documentation.

Availability

DeepSeek V4.1-Flash-1017 is available through DeepSeek's website and app. API users can access the updated model with the identifier deepseek-flash, without changing their existing integration.The model weights are also available for download and deployment through Hugging Face.DeepSeek V4.1 Flash is additionally listed in the BotHub model catalog, where it can be accessed through a unified API alongside models from other providers.