DeepSeek ships V4.1-Flash-1017 with open MIT weights
The 552B-parameter model goes live on DeepSeek's site, app and API the same day, scoring above Claude Opus 5 on several agentic benchmarks.
DeepSeek released the AI model DeepSeek-V4.1-Flash-1017 on October 17, 2026 (Japan time), according to PC Watch. The open weights were published for free on Hugging Face under the MIT license the same day, with commercial use permitted, and the model went live on the company's website, app and API.
The model has 552B total parameters and scored 77.0 on the coding benchmark DeepSWE v1.1, above Claude Opus 5's 74.0. It also exceeded Anthropic's model on Automation-Bench v1.0.6 (60.5 vs 50.3), Agents' Last Exam (35.4 vs 28.6) and Apex Bench v1.1 (72.1 vs 65.8).
Not every comparison favoured DeepSeek. The new model scored 31.2 to 42.3 on Terminal-Bench 4.0, below Claude Opus 5's 51.8, and 42.6 against 56.3 on HLE.
DeepSeek-V4.1-Flash-1017 accepts text and images, has a 1-million-token context length and recommends allocating at least 256,000 output tokens.
The company also opened its Search API and Fetch API to all users at $0.75 per 1,000 calls each.
A minor update to September's model
DeepSeek-V4.1-Flash-1017 is a minor update of DeepSeek-V4.1-Flash, which DeepSeek released in September 2026. The new version arrived only 37 days after the previous one.
Quick answers
Is DeepSeek-V4.1-Flash-1017 free to use commercially?
Yes. Its open weights were published for free on Hugging Face under the MIT license, with commercial use permitted.
How large is DeepSeek-V4.1-Flash-1017's context window?
The model has a 1-million-token context length and recommends allocating at least 256,000 output tokens.
How much do DeepSeek's Search API and Fetch API cost?
Both APIs were opened to all users at $0.75 per 1,000 calls each.