
Tech • IA • Crypto
Z.AI’s GLM 5.2 emerges as a near-peer to top Western systems, matching Claude Opus on coding and vulnerability detection benchmarks. The model reportedly reaches around 74.4% vs 75.1% on advanced tests, signaling narrowing gaps. Its release in June 2026 has triggered scrutiny from U.S. authorities. The development reframes AI competition as a fast-moving geopolitical race.
GLM 5.2 is distributed under a permissive MIT license, enabling unrestricted download and modification. This open-weight approach boosts enterprise autonomy but weakens centralized safeguards. Cybersecurity experts warn it could facilitate off-grid misuse and harder-to-detect attacks. The debate revives tensions between openness and control in advanced AI.
GLM 5.2 is touted as up to 85% cheaper than rivals like ChatGPT 5.5, largely due to token efficiency. However, higher compute demands complicate the real economics. Industry focus is shifting toward cost per completed task, not per-token pricing. Early analyses suggest the advantage may narrow under real workloads.
Anthropic introduces Claude Sonnet 5, delivering near-premium performance at mid-tier pricing. It rivals Opus 4.8 on reasoning and coding while costing about $2 input / $10 output per million tokens. The model becomes the default for free users, raising the baseline across the market. This signals aggressive competition on both capability and accessibility.
Sonnet 5 expands agentic behavior with browser use, terminal control, and iterative task execution. It reliably completes multi-step workflows that earlier models often failed to finish. Benchmarks show 80%+ performance on agentic coding scenarios. This marks a shift from chatbots to semi-autonomous digital workers.
Claude Fable 5 returns globally after a brief withdrawal tied to U.S. government concerns. Officials pushed for stronger protections against jailbreaks and misuse. Anthropic says it added improved classifiers and monitoring systems. The episode highlights growing state influence over frontier model deployment.
OpenAI leadership warns that AI capable of self-directed research may arrive sooner than expected. Systems are increasingly able to generate hypotheses, run experiments, and iterate with minimal oversight. Progress continues across scaling, reasoning, and long-horizon tasks despite skepticism. The shift raises urgency around evaluation and safety frameworks.
Rocket Lab will acquire Iridium Communications for $8 billion, a 20% premium deal. The move adds a 66-satellite network and valuable spectrum assets. It accelerates vertical integration, positioning Rocket Lab against SpaceX Starlink. The deal reflects consolidation as space firms race to control end-to-end infrastructure.