MiniMax and DeepSeek Release New Models, US DHS Orders Removal of DeepSeek Tech

Large model vendors continue updating their model matrices and refreshing open-source benchmarks, while federal system security compliance and agent interception become industry focuses.

The essentials

MiniMax Releases M3.1 Flash

The official announcement states that M3.1 Flash achieves frontier-level intelligence at one-third the price and about 3 times the speed, with an Artificial Analysis intelligence index of 42, a cost per single task of about $0.22, and 428B weights released on October 23.

Source ↗

US DHS Orders Removal of DeepSeek

The US Department of Homeland Security issued an order based on supply chain laws, requiring all federal civilian agencies to immediately stop procurement and remove DeepSeek and its affiliated company products from IT systems, covering related third-party services.

Source ↗

fal and BFL Release FLUX 3 Max Video Model

BFL and fal partnered to launch FLUX 3 Max. The official announcement states it is 22 times faster than Seedance 2.5 at equivalent quality, generating 10 seconds of video in 12 seconds, tying for first place on the image-to-video leaderboard.

Source ↗

In brief

DeepSeek V4.1 Flash 1017 Released

Artificial Analysis records an intelligence index of 45, 6 points higher than the previous generation, tying for second place in open-source weights.

Source ↗

Inception Releases Mercury 3 Diffusion Model

Generation speed remains above 1000 tokens/s, supporting 1M context, and launched on OpenRouter.

Source ↗

Hackers Use AI Agents to Breach Over 100 Websites

Reuters reported that hackers used AI agents to breach over 100 company websites, with an operation cost of about $5000.

Source ↗

Major Platforms Begin Blocking Agent Access

Amazon, Delta, and JPMorgan Chase are blocking AI agents from accessing their systems.

Anthropic Preparing New Model

Anonymous testers state that reasoning and intuition are significantly stronger than currently commercialized Anthropic models.

Ant inclusionAI Open-Sources Speech Model

Released Ming-3.0-tts speech synthesis model on GitHub, open-sourced under the MIT license.

Source ↗