DeepSeek V4.1-Flash-1017: Higher Metrics, Search and Fetch APIs...

DeepSeek V4.1-Flash-1017: Higher Metrics, Search and Fetch APIs Open to All

DeepSeek released a minor checkpoint, DeepSeek-V4.1-Flash-1017, with open weights (MIT), surpassing the previous Flash on all 11 stated benchmarks, and for the first time opened Search and Fetch APIs on its own search engine at $0.75 per 1,000 successful calls.

Compiled by Sergey KostenchukPublished 2026-10-17Updated 2026-10-17

2026-10-17 Research Anthropic

🐋 DeepSeek V4.1-Flash-1017: Higher Metrics, Search and Fetch APIs for All

DeepSeek released a minor V4.1-Flash checkpoint from October 17, 2026 — DeepSeek-V4.1-Flash-1017 with open weights (MIT). Compared to the previous Flash — a leap: Terminal-Bench 4.0 from 31.2 to 42.3, ProgramBench from 20.3 to 29.1. Compared to Claude Opus 5, the results are mixed; the figures are self-reported.

🌍 DeepSeek has for the first time opened its own search engine as a cheap Search and Fetch API for agents and RAG: $0.75 per 1,000 successful calls for each endpoint, free in Claude Code, Codex, and DeepSeek Harness until October 31.

👤 The model is available via API as deepseek-flash ($0.30 per 1M input tokens and $1.20 per 1M output tokens) and with open weights on Hugging Face. Until October 31, the Search/Fetch APIs can be tested for free — a ready-made web tool for your own agents.

Source 1: https://api-docs.deepseek.com/news/news261017/ Source 2: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017

Sources