PyyanThe AI Index

Pyyan / News / 17 October 2026

Open weightsDeepSeek

DeepSeek refreshed V4.1-Flash and kept the MIT licence

11 of 11benchmarks improved, by 2.0 to 11.1 points

DeepSeek published V4.1-Flash-1017 on Hugging Face under the MIT licence. It is an update to V4.1 Flash rather than a new model. DeepSeek reports beating the previous version on all eleven benchmarks it published, by 2.0 to 11.1 points, and one outlet puts it at 77.0 on DeepSWE v1.1 against 74.0 for Claude Opus 5. Context stays at 1M tokens. Parameter counts differ: 763 billion on the Hub, 552 billion described as the backbone elsewhere. The checkpoint is about 510GB.

DeepSeekMarkTechPostfrom the source itself

Related

Open weightsTwo days after being named by US agencies, DeepSeek released a bigger open model552B parameters, MIT licence, 1M contextPolicyAltman, Amodei, Bengio and Delangue briefed the UN Security Council on AI riskFirst high-level Security Council briefing devoted to AI safetySafetyOne attacker used hundreds of AI agents to break into 395 organisations, and the agents ignored its rules26 seconds to compromise 11 organisations

← All the news, newest first