ENFR
8news

Tech • IA • Crypto

TodayTopicsVideosCryptoArchivesFavorites

GLM 5.2 vs Claude, Rocket Lab $8B Iridium, MetaMP

AIThursday, July 2, 2026· 11 videos

Briefing

Audio player
0:00 / 0:00

GLM 5.2 challenges Western leaders

Z.ai’s GLM 5.2 reaches near parity with top models, scoring about 74.4% on coding benchmarks versus Claude Opus at 75.1%. The model supports extended, hour-long workflows, signaling stronger applied reasoning. Its release under an MIT license allows unrestricted global use and modification. The combination of performance and openness intensifies competitive pressure on closed Western systems.

GLM 5.2 slashes AI costs

GLM 5.2 is reported to be 85% cheaper than comparable systems like ChatGPT 5.5, driven by efficiency and lower token usage. This pricing shift could reshape enterprise adoption and margins across the industry. The model was reportedly trained on Huawei Ascend 910B chips instead of Nvidia, raising questions about shifting hardware dependencies. If validated, this signals a major geopolitical and supply-chain inflection.

Rocket Lab acquires Iridium for $8B

Rocket Lab is buying Iridium Communications in a deal valued at $8 billion, including a 20% premium. Iridium operates a 66-satellite LEO constellation serving enterprise, maritime, and government clients. The acquisition marks one of the largest consolidations in commercial space in years. It gives Rocket Lab immediate access to spectrum, infrastructure, and recurring connectivity revenue.

Rocket Lab targets SpaceX Starlink

The deal transforms Rocket Lab into a vertically integrated space company spanning launch, manufacturing, and services. This positions it more directly against SpaceX’s Starlink, which operates around 10,000 satellites. While smaller, Iridium’s established network and customer base provide a strategic foothold. The move signals intensifying competition in satellite connectivity markets.

Claude Sonnet 5 boosts autonomy

Anthropic’s Claude Sonnet 5 introduces stronger agentic capabilities, including browser use, terminal control, and iterative self-correction. It achieves over 80% in agentic coding benchmarks and handles multi-step workflows more reliably. The model is now default for free users, significantly raising baseline AI capability. This marks a shift toward practical semi-autonomous assistants.

Sonnet 5 raises cost tradeoffs

Claude Sonnet 5 improves output quality but uses 30–35% more reasoning tokens, increasing operational costs. High-effort runs can exceed $3 per task, compared to roughly $0.50 previously. Medium-effort settings offer a better cost-performance balance for most use cases. The release highlights diminishing returns as models approach top-tier performance.

Meta pushes MetaMP compute platform

Meta Platforms is preparing MetaMP, a service offering AI models and compute capacity to external customers. The move targets competitors like AWS, Google Cloud, and Microsoft Azure. It follows hundreds of billions of dollars in infrastructure investment. The strategy aims to monetize excess capacity while expanding Meta’s role in AI infrastructure.

OpenAI signals autonomous research leap

OpenAI leadership says AI is approaching self-sustaining research systems capable of generating hypotheses and running experiments. Progress continues across scaling laws, spanning pre-training to long-horizon reasoning. Examples echo breakthroughs like AlphaGo’s Move 37, now appearing in science and coding. The shift raises urgency around safety, evaluation, and governance.

Videos covered

Previous briefings · AI