// STORY_01
GLM-5.2 Captures 40% of Developer Tokens, Trails Claude Opus by 1 Point on FrontierSWE
Z.ai's open-weight GLM-5.2 now accounts for 40% of developer tokens on major coding platforms and scored 74.4 on FrontierSWE (independently evaluated by Proximal), within 1 point of Claude Opus 4.8's 75.1 and ahead of GPT-5.5. On Terminal-Bench 2.1 it scored 81.0, up from GLM-5.1's 63.5, while Humanity's Last Exam (with tools) hit 54.7, ahead of GPT-5.5's 52.2.
// STORY_02
US Weighs New Framework for Releasing Open-Source AI Models as Chinese Rivals Gain Ground
The Trump administration is discussing a new framework for releasing advanced open-source AI models as businesses increasingly adopt low-cost Chinese AI systems. The policy debate centers on balancing national security concerns with the competitive pressure from cheaper open-weight alternatives flooding the market.
// STORY_03
US Companies Ditch American AI Models for Cheaper Chinese Alternatives
According to OpenRouter data, US enterprises are increasingly routing traffic to Chinese open-weight models, with Featherless AI CEO Eugene Cheah calling Chinese models "the elephant in the room." Databricks' enterprise benchmark found GLM-5.2 matches expensive frontier models on everyday coding tasks at roughly two-thirds the cost, signaling a shift from leaderboard rankings to real-world cost-performance.
// STORY_04
GPT-5.6 Sol Takes #1 on Design Arena, Beating Claude Fable 5 with Elo 1353
OpenAI's GPT-5.6 Sol secured the top spot on the Design Arena leaderboard with an Elo score of 1353, narrowly beating Claude Fable 5. On the Artificial Analysis Intelligence Index, Sol placed second overall but completed tasks 61% faster and at half the cost of its closest rival. OpenAI president Greg Brockman called it a "big milestone."
// STORY_05
Meta's In-House AI Chip "Iris" Enters Production in September, Targets 7 GW Infrastructure
Meta will begin manufacturing its in-house AI chip, code-named Iris, in September, designed with Broadcom and built by TSMC. Meta plans to deploy 7 gigawatts of computing infrastructure this year and release a new chip every 6 months through 2027, up from the typical yearly cadence. Total AI infrastructure spend could reach $145 billion.
// STORY_06
China's Optical Chip Breakthrough Boosts AI Speed 100-Fold with Fraction of Compute Power
Peking University researchers developed an all-optical interconnect system that links standard electronic chips with specific algorithms, achieving 100-fold AI speed improvements using a fraction of the compute power. The breakthrough could reshape AI hardware architecture by bypassing traditional electronic bottlenecks.
// STORY_07
Intel Commits EUR 5 Billion to Ireland AI Chip Manufacturing Expansion
Intel announced a EUR 5 billion investment to modernize its Leixlip, Ireland fabrication facility, Europe's most sophisticated semiconductor production complex. The expansion will increase manufacturing volume for Intel Xeon 6 processors and upcoming Xeon chips, introducing cutting-edge manufacturing technology to meet AI-driven demand.