Model access updates and AI safety #83

July 20, 2026 — 2 min read

Today's Letter

The provided primary text is a third-party guide for setting up a spare Mac for Claude Code remote control
According to the factcheck result, auxiliary sources discuss different topics and do not confirm the setup details described here
No official Anthropic source is included in the materials for the claimed configuration, capability, or security guidance
The available sources are not sufficient to turn this item into a verified product or platform update
This item is not suitable for newsletter body use without stronger source alignment or official confirmation

Source: ykdojo.github.io
More: simonwillison.net · news.hada.io · news.aibase.com

Qwen Cloud updated its Token Plan and now lists Qwen3.8-Max-Preview as part of the plan's advanced model access
The company says one subscription provides unified access across text, vision, speech, and image generation models
The model lineup shown on the pricing page also includes glm-5.2, deepseek-v4-pro, and wan2.7-image-pro
Individual plans are now available, while the company says Team pricing has been reduced and overall pricing is around 40% lower
The Lite tier supports access to multiple models, harness tools, and 1-2 concurrent agents
The Standard tier offers 4x the Lite credits and raises concurrency to 3-4 agents
The Pro tier offers 16x the Lite credits and raises concurrency to 6-8 agents
Qwen Cloud says the plan works with tools using OpenAI and Anthropic protocols, including Codex, Claude Code, Cursor, Cline, and OpenCode

Source: qwencloud.com
More: the-decoder.com · scmp.com · officechai.com

Google DeepMind and Isomorphic Labs published a joint bioresilience framework on July 16, 2026, focused on preventing model misuse and supporting outbreak preparedness
The program says more than 15 partnerships were advanced over the past 12 months with government bodies, biosecurity organizations, and research groups
The prevention track applies a four-step safety process for models such as Gemini: threat modeling, evaluations, mitigations, and monitoring
The companies said they are adapting SynthID watermarking for biology so DNA synthesis providers could screen potentially risky AI-generated biological sequences
The detection track highlights AlphaEvolve for optimizing metagenomic sequencing workflows to make pathogen surveillance faster, more accurate, and lower cost
Google DeepMind also pointed to AlphaGenome and protein function annotation work as tools for identifying novel patterns and emerging threats from sequence data
The response track builds on AlphaFold and grants trusted researchers access to newer AI systems to accelerate vaccine and countermeasure design
The post positions AlphaFold, AlphaGenome, and Isomorphic Labs' AI-powered Drug Design Engine as core systems for prevention, detection, and response use cases

Source: deepmind.google
More: artificialintelligence-news.com

Jocoletter curates AI, software, and product trends for developers and builders.

#Anthropic #GoogleDeepMind #QwenCloud