China's Open-Weight Models Keep Closing the Gap — DeepSeek V4 Pro and GLM 5.3 Both Ship Under Permissive Licenses

DeepSeek V4 Pro landed as the second most intelligent open-weights model benchmarked, while GLM 5.3 topped a security-flaw benchmark — and both plan to give the weights away.

Two Chinese labs made moves this week that should command attention from anyone building on open weights. DeepSeek V4 Pro 0813 scored 53 on the Artificial Analysis Intelligence Index, and — critically — DeepSeek published the weights under the MIT license. As @ArtificialAnlys put it, that makes it "the second most intelligent open weights model we have benchmarked." MIT licensing is about as permissive as it gets, which means the model is not just viewable but genuinely usable in commercial products without the strings attached to more restrictive open releases.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.