🔥 OpenAI Previews GPT-5.6 Sol

PLUS : GLM-5.2 Beats GPT-5.5 in Coding

Author

The Summary AI
June 28, 2026

In partnership with

Welcome back!

The next frontier of AI has arrived, but we can’t use it. OpenAI has unveiled GPT-5.6 Sol, alongside two lower-cost variants. But, just like Claude Fable 5, the release is on hold pending US government cybersecurity approval, raising new questions about who gets access to the most advanced models. Let’s unpack…

Today’s Summary:

  • 🔥 OpenAI launches GPT-5.6 under US Gov oversight

  • 🚀 GLM-5.2 frontier open source AI

  • 📜 AI reads ancient sealed scroll

  • 👀 Anthropic Fable 5 nears return

  • 📄 Mistral launches OCR 4

  • 💻 Gemini Flash gains computer control

  • 🛠️ 2 new tools

TOP STORY

OpenAI launches GPT-5.6 Sol Preview under US government oversight

The Summary: OpenAI announced GPT-5.6 Sol, its new frontier model, alongside two cheaper siblings: Terra (half the price of GPT-5.5 with similar performance) and Luna (low-cost tier). The US government required OpenAI to begin with a limited rollout to only 20 trusted partners, while conditions for public launch get sorted out, likely by mid-to-late July.

Key details:

  • Scores 88.8% on TerminalBench 2.1 (agentic coding), edging out Claude Mythos 5 at 88%; the "ultra" multi-agent mode hits 91.9%

  • A Cerebras integration in July will push speeds to 750 tokens/sec

  • Priced the same as GPT-5.5, but the token efficiency gains could make real-world task costs meaningfully lower

  • All three GPT-5.6 models crossed OpenAI's internal "High" cyber risk threshold

Why it matters: The most capable AI models in existence, Anthropic Fable 5 and Mythos 5, and now OpenAI Sol, Terra and Luna, are locked from general release, pending US government cybersecurity review. OpenAI criticized the current process, calling it unsustainable and asking for a clear release framework. The new policy changes the competitive landscape in ways that go beyond a delayed launch. The question is no longer which model is best, but which one we are actually allowed to use.

FROM OUR PARTNERS

The Agent Pack Marketers Can Use Today

22 ChatGPT Agents Built for Every Marketing Job

Most marketers use ChatGPT to do general research and then call it an AI strategy. The ones outperforming them are deploying specialized agents built for specific jobs.

We put together 22 plug-and-play ChatGPT marketing agents that handle the work eating your week, each with built-in instructions and structured outputs ready to go in under 5 minutes.

Subscribe to Marketing Against the Grain and get all 22 free.

Inside you'll find:

  • Competitive intelligence agent that visits competitor websites and builds detailed comparison matrices automatically

  • Customer feedback analyzer that ranks improvement opportunities by business impact

  • Social listening specialist that monitors brand mentions and flags reputation risks before they escalate

  • Campaign optimization agents that handle attribution analysis and surface what is actually driving results

Your competitors are already running agents like these.

Get 22 ChatGPT Marketing Agents free when you subscribe to Marketing Against the Grain today.

Z.AI

GLM-5.2 pushes open source into frontier territory

The Summary: Z.ai has released GLM-5.2, an open-weight 744B-parameter Mixture-of-Experts model with 1M context and an MIT license. It is now the best open-weight model in benchmarks and beats GPT-5.5 on several software engineering benchmarks while keeping API pricing far below proprietary levels.

Key details:

  • Ranks as the highest-scoring open-weight model on the Artificial Analysis Intelligence Index

  • Uses an MIT license, allowing unrestricted commercial deployment, local hosting, and fine-tuning without usage restrictions or regional limits

  • Running it locally requires 256GB RAM minimum (quantized versions) and 24GB VRAM for MoE offloading

  • For now, it lacks vision, meaning it cannot see input images

Why it matters: The gap between open-weights and proprietary models has historically been measured in months. GLM-5.2 suggests that gap has collapsed to near zero for coding. The one open question is reasoning efficiency, where GLM-5.2 uses 43K output tokens per task on average, nearly 3x as many as GPT-5.5, which means tasks may be slower for production pipelines.

FROM OUR PARTNERS

Upgrade Your AI Skills After Hours

Free Weekly AI Sessions for Experienced Software Engineers.

Every Wednesday at 5 PM CT, Gauntlet AI professors teach a live, hands-on AI engineering session — completely free. If you're nontechnical, this isn't for you. New topic every week, built for engineers who want to build, not just watch. See upcoming sessions.

AI + HISTORY

AI deciphers a charred 2,000-year-old scroll without opening it

The Summary: Researchers used AI to read a carbonized Roman scroll without physically unrolling it. High-resolution X-ray scans from the world’s most powerful synchrotron beamline, paired with specialized AI models, recovered 1.4 meters of ancient Greek text hidden since the eruption of Mount Vesuvius buried the library in 79 AD.

Key details:

  • PHerc. 1667 contains about 22 columns of Greek text and is the first Herculaneum scroll read from beginning to end

  • The recovered work is a Stoic treatise from the 2nd century BC, naming Aristocreon, the nephew and student of Chrysippus

  • Around 600 unopened Herculaneum scrolls exist, and 30 have been scanned. One scan produces 260 terabytes of data.

Why it matters: Only 1% of ancient Greek and Latin writing survived, hand-picked by medieval monks. The Herculaneum scrolls were buried and bypassed this filter entirely. AI just read a scroll burned in 79 AD without touching a single fiber of it. Estimates put 10,000+ scrolls still underground in the Herculaneum site. This could completely transform our understanding of ancient history.

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/