Alibaba launches Qwen 3.8-Max, 2.4T params challenge global frontierMODEL
Alibaba officially released Qwen 3.8-Max on Aug 3, with 2.4 trillion total parameters (MoE) and 95B active, excelling at coding, office work, research and long-horizon end-to-end tasks. On Arena it ranks just behind Claude series, delivering 265 code commits in 16 days of independent programming. Legal review flagged 1,284 clauses in one hour; chip design sandbox reduced gate count from 8,298 to 678 in 500 rounds. API priced at $2/$6 per million I/O tokens, undercutting Kimi K3. Weights open-source next week.[1]

On the AI side: the contradiction between frontier model capability leaps and safety governance is reaching fever pitch. Alibaba's Qwen 3.8-Max with 2.4T params challenges the global frontier, DeepSeek's Agent capability jumps 6x, OpenAI Astra cracks 10 math problems — the capability race is still accelerating. Yet simultaneously, Altman pivoting from acceleration to "deceleration", the White House summoning closed-door security talks, 15 state AGs investigating the agent hack, and Google AI fixing more bugs in a month than two years — the urgency of safety governance is rising like never before. Palantir's 93% revenue surge reveals the explosive emergence of "AI sovereignty" as an entirely new demand dimension.
On the Web3 side: regulatory uncertainty and security incidents deliver a double blow, pushing sentiment to extreme fear. CLARITY Act pass odds fall to 31%, Coldcard vulnerability exceeds $114M in losses, Hashdex liquidates its Bitcoin ETF, FalconX cuts 10% — multiple negatives stacking up. But structural trends like Mastercard acquiring BVNK, BlackRock launching tokenized funds, and AI agent DeFi growth keep advancing — institutional infrastructure and tech innovation don't stop for the bear.