New AI Models from OpenAI, xAI, and Meta Spark Competition; EU Chat Control Faces Criticism

Key Takeaways

1. OpenAI launched GPT-5.6 Sol, achieving SOTA on ARC-AGI-3 (7.8%) and Terminal-Bench 2.1 (88.8%), with a focus on enterprise tasks and agentic workflows.
2. xAI’s Grok 4.5 outperforms competitors across multiple benchmarks and demonstrates strong agentic/coding abilities, with plans for 1M context, 2T parameter model, and monthly releases.
3. Meta’s Muse Spark 1.1 shows strong agentic performance with 1M token context and parallel sub-agent execution, challenging GPT-5.6 and Grok.
4. EU passed the Chat Control bill, allowing warrantless scanning of private messages, drawing sharp criticism from tech figures.
5. John Collison highlights that solo founders earning >$1M/year have tripled since 2021, underscoring the viability of solo entrepreneurship.

1. AI Model Wars: GPT-5.6 Sol, Grok 4.5, and Muse Spark 1.1

1. OpenAI released GPT-5.6 Sol, described as its best model, achieving 88.8% on Terminal-Bench 2.1 (91.9% in ultra mode), surpassing prior models and competitors. It also scored 7.8% on ARC-AGI-3, the first verified frontier model to beat that game. Sol powers ChatGPT Work, Codex, and a new desktop app, with Terra and Luna further improving efficiency. — via 1 2 3
2. xAI’s Grok 4.5 leads in multiple benchmarks (AutomationBench-AA, Terminal-Bench v2, Harvey Legal, SWE Marathon, etc.), outperforming GPT-5.6 and Claude on professional work benchmarks. It demonstrated one‑hour creation of an FPS game and autonomous 3D scene construction in UE5.8. Future plans include expanding context to 1M tokens, a 2T parameter base model, monthly release cadence, and hardware‑accelerated inference. — via 1 2 3
3. Meta’s Muse Spark 1.1 received praise for agentic performance, tool use, and computer use, with a 1M token context window and parallel sub‑agent execution. It is now available on AI Gateway at Vercel. Garry Tan noted its strong showing on OpenClaw. — via 1 2 3
4. The rapid model releases lead Guillermo Rauch to call this “model release week,” noting that Meta Spark 1.1, Grok 4.5, and GLM 5.2 are capturing token market share from GPT‑5.6. Amjad Masad adds that the LLM market is no longer a monopoly threat, with multiple players advancing quickly. — via 1 2

2. Agentic Tooling and Platform Updates

1. OpenAI’s ChatGPT Work, built on Codex, enables long‑running projects across apps and files. Sam Altman emphasized Codex remains core and will not disappear. GPT‑5.6 also powers Microsoft 365 Copilot. — via 1 2
2. Perplexity Computer added Analytics to track model spend, and Aravind Srinivas noted the availability of GPT‑5.6 Sol, GLM 5.2, and Claude Opus 4.8 Fast Mode as orchestrators, with Sol and Terra serving as search models. — via 1 2
3. Vercel’s v0 now allows generating designs, prototypes, and full‑stack apps with custom design systems. Guillermo Rauch also announced zero‑config import of Lovable apps to Vercel, leveraging the Nitro runtime. — via 1 2
4. 37signals’ BC5 introduced a new feature called Wormholes, demonstrated via video. Jason Fried also emphasized building for small teams over enterprise logos. — via 1
5. Jack Dorsey shared principles for AI agents: they should request minimal information rather than full identity, and interaction is shifting from commanding to asking for suggestions. — via 1 2

3. Regulation, Politics, and Social Impact

1. The EU passed the Chat Control bill, which allows scanning private messages, emails, and photos without a warrant. Paul Graham criticized the undemocratic process (voted while most MEPs were on break). @levelsio called it totalitarian. — via 1 2
2. Paul Graham reported that ICE shot and killed a worker who had lived in the US for 35 years, disputing the official claim that the worker tried to ram a vehicle. Video evidence suggests ICE chased and stopped the worker. — via 1
3. Garry Tan warned of coordinated information warfare by China, Russia, and Iran using state media to turn data center controversies into internal US division. — via 1

4. Entrepreneurship and Market Data

1. John Collison shared that the percentage of solo founders earning over $1M per year more than tripled from 2021 to 2025, signaling a shift toward viable solo entrepreneurship. — via 1
2. Ryan Hoover introduced Roll Up Vehicles and the Value Add Network, which connects founders with high‑quality angels and VCs to complete funding rounds. — via 1
3. Amjad Masad observed that AI makes coding more flexible but runtimes more rigid, pushing infrastructure teams to write formal specs. — via 1