Executive Summary
- OpenAI has severed API model access for coding platform Cursor following its acquisition by SpaceXAI, escalating corporate rivalry with Elon Musk and exposing ecosystem reliance risks.
- Anthropic integrated cross-session memory across Claude Chat and Claude Cowork across all devices, alongside embedding a native browser agent within its Cowork environment.
- Anthropic confirmed a permanent 25% uplift to Claude Code weekly usage thresholds for paid subscribers starting 14 September, expanding developer output limits.
- Tencent released its open-weight Hunyuan Hy4 preview model under an Apache 2.0 licence, boasting a 770B parameter Mixture-of-Experts architecture and 1M token context.
- OpenAI revealed plans to integrate sponsored advertisements within ChatGPT's free and 'Go' tiers, marking a major shift towards ad-supported monetization.
In Detail
Anthropic
- Claude Cowork unifies memory with Chat across all devices
- Claude Cowork receives a built-in browser for autonomous web navigation
- Claude Code weekly limits increased by 25% for paid plans starting Sept 14
Reuters / Engadget
- OpenAI terminates model access for SpaceX-owned Cursor editor due to TOS concerns with SpaceXAI
Hugging Face / Tech Media
- Tencent ships Hunyuan Hy4 preview with 770B MoE parameters and 1M context under Apache 2.0
- Z.ai releases open-weight GLM-5.3 with custom license requiring security reviews for hyperscalers
Google DeepMind
- Gemini 3.7 Flash released for complex agentic tasks and software engineering
- Gemini Omni 1.1 Flash launched supporting 40-second scenes and 4K video generation
- Google Assistant scheduled for full retirement starting September 2026 as Gemini takes over
Igor's Lab / Inc.
- OpenAI to introduce ads into ChatGPT Free and Go tiers
- DALL-E GPT official retirement scheduled for August 30
Model Releases
- Hunyuan Hy4 Preview — Tencent. Tencent released an open-weight 770-billion parameter mixture-of-experts (MoE) model under the Apache 2.0 license featuring a 1-million token context window. Read more
- Gemini 3.7 Flash — Google DeepMind. Google released Gemini 3.7 Flash, optimized for high-speed agentic software engineering and multi-step reasoning tasks at Flash-tier latency. Read more
- Gemini Omni 1.1 Flash — Google DeepMind. Updated multimodal generation model capable of producing 40-second continuous 4K video scenes. Read more
What to Watch
Interpretation, not reported fact. Each point follows from a source listed below — follow the links and judge for yourself.
- Developer tool API access may become increasingly tied to corporate alliances and proprietary ecosystems. (Basis: OpenAI cutting off model access to Cursor following its acquisition by SpaceXAI.)
- Enterprise AI workspace products are rapidly adopting unified state and native web navigation as baseline agentic capabilities. (Basis: Anthropic unifying memory across Chat and Cowork alongside launching a built-in browser for agentic web tasks.)
Sources
- Anthropic (https://www.anthropic.com)
- Reuters / Engadget (https://www.reuters.com)
- Hugging Face / Tech Media (https://huggingface.co)
- Google DeepMind (https://deepmind.google)
- Igor's Lab / Inc. (https://www.inc.com)