AI Morning Brief — 13 July 2026
GPT-5.6 Sol claimed the top position on Design Arena overnight, edging past Claude Fable 5 on frontend design benchmarks and prompting Anthropic to extend Fable access for paid subscribers for a second week running. The session also brought Grok 4.5 claims of Opus-class browser performance, Apple filing a trade-secret suit against OpenAI, and Anthropic shipping a native browser inside Claude Code.
Top stories
- GPT-5.6 Sol leads Design Arena. Official results put GPT-5.6 Sol first overall with an Elo of 1,353, placing it above Claude Fable 5 and in the same performance band as GLM 5.2 on frontend design tasks — an 18-position and 60-point Elo jump in a single cycle. via X · @gdb
- Anthropic extends Fable 5 access, again. Following the benchmark result, Anthropic extended Claude Fable 5 availability on all paid plans and kept Claude Code weekly rate limits 50% higher, through July 19 — the second such extension in recent weeks amid competitive pressure from GPT-5.6 Sol. via Simon Willison
- Grok 4.5 claims Opus-class browser performance. Third-party evaluators placed Grok 4.5 just below Claude Opus and marginally above GPT-5.6-Sol for browser use, at roughly 10% lower cost than Opus. xAI also claimed Grok 4.5 ranks above Fable on some software benchmarks. via X · @elonmusk
- Claude Code ships a built-in browser. Anthropic added a native browser to Claude Code that allows the agent to read, click, and type on external websites directly inside the development environment. Write actions on third-party sites are screened by classifiers; purchases and account creations require explicit user approval. via THE DECODER
- Apple sues OpenAI for trade secret theft. Apple filed a lawsuit alleging OpenAI stole trade secrets “at every level,” with claims linked to former Apple employees. The suit adds legal friction to a relationship that has otherwise been moving toward partnership. via Business Insider
- Meta withdraws Muse Image non-consent feature. Days after announcing a feature that let any user generate AI photos of another person by tagging their public Instagram handle, Meta pulled it following widespread criticism, acknowledging it “missed the mark.” via THE DECODER
Who shipped
Anthropic drove the overnight cycle on two fronts: a native browser capability for Claude Code and a further extension of Fable 5 availability under competitive pressure from GPT-5.6 Sol. xAI positioned Grok 4.5 against frontier-tier rivals through third-party browser-use and software benchmarks. OpenAI did not announce new releases within the window but dominated discussion through the Design Arena result and the Apple trade-secret lawsuit.
Open-source pulse
Soofi S 30B-A3B, a sovereign Mixture-of-Experts model for German and English trained on Deutsche Telekom infrastructure in Munich, published its full pretraining tech report. Xiaomi quietly uploaded MiMo-V2.5-DFlash weights — 300B+ parameters — to Hugging Face without a formal announcement. llama.cpp shipped thirteen releases overnight (b9970–b9982), covering DeepSeek V4 cache fixes, agentic checkpoint corrections, and Vulkan routing patches for Adreno GPUs.
Money, infra & hardware
S&P Global downgraded Oracle‘s credit rating to BBB- — one notch above junk — citing OpenAI as a key credit risk; OpenAI accounts for roughly half of Oracle’s $638 billion in contractual obligations. Separately, Irish data centres now consume 23% of the country’s national electricity supply, via The Register, underscoring the continued strain AI infrastructure places on national grids.
Quiet corners
Google DeepMind, Mistral, and Cohere were absent from the overnight cycle. Meta appeared only in connection with withdrawing a product feature rather than shipping one.
By the numbers
- 217 stories in 24 h across 20+ sources
- Most-mentioned model: Claude / Fable
- Most-mentioned lab: Anthropic
- Notable absences: Google DeepMind, Mistral, Cohere
Compiled by AI Feed’s editor from all 217 headlines published on 13 July 2026.