🔒 OpenAI Launches New GPT-5.6 for a Chosen Few

OpenAI just built its most capable model yet, and almost no one can use it. GPT-5.6 launched this week as a limited preview to a small group of vetted partners, with the U.S. government getting an early look before the rest of us. The reason sits in a quiet shift that's been building all year. Washington now has a say in how the most powerful AI reaches the public, and this is the clearest sign yet of what that looks like in practice. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: OpenAI is rolling out its new GPT-5.6 to only a handful of firms with Washington in the loop.

GPT-5.6 arrives in three tiers, with Sol as the flagship, Terra priced at about half of GPT-5.5, and Luna lowest at a dollar per million input tokens. Sol set a new high on the Terminal-Bench 2.1 coding test and now offers a max mode for deeper reasoning and an ultra mode that splits work across subagents. Before release OpenAI ran more than 700,000 GPU hours of automated red-teaming, and a faster version arrives on Cerebras chips at up to 750 tokens per second in July. Even with that jump in capability, Sol landed under the company's own Cyber Critical safety line.

A June executive order gave Washington a standing role in how frontier models reach the public, and GPT-5.6 is the first big launch to run through it. The government can now preview these models early and help pick who gets them first. For everyday developers and ChatGPT users, the most capable AI now arrives on a delay set by a process they never see. The people building these tools and the people allowed to use them are drifting further apart.

Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.

ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.

Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.

The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.

Start building free

Evolving AI: Washington cleared Anthropic to send Mythos 5 back to 100 vetted firms.

Washington approved the redeployment after deciding that appropriate safeguards were in place, capping two weeks of daily talks that began with the June 12 shutdown. The cleared list reaches Fortune 500 companies and federal agencies that guard the country's infrastructure. Those organizations now get a model that can hunt software flaws on its own across the systems they defend.

Mythos 5 puts that same skill to work on defense, scanning the networks behind power grids and hospitals faster than human teams could. The payoff reaches people as a billing system that stays up or a clinic that keeps running, protection that works out of sight in the services daily life depends on.

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.

Browse Free Sessions

Evolving AI: Musk's $60B Cursor bet is now rivaling Opus in early private evaluations.

Grok 4.5 got its coding edge from Cursor after SpaceX agreed to buy maker Anysphere for 60 billion dollars in mid June. The model runs on xAI's V9 foundation which finished training on May 26, about triple the size of the version still handling Grok on X. That acquisition closes later this year and pulls a leading coding tool directly into Musk's own pipeline.

Musk is keeping Grok 4.5 inside his companies for now, so most readers cannot try it or check the Opus claim for themselves. What reaches them sooner is the release pace, since a new model each month sets how fast the tools they will eventually use keep moving while this beta stays out of reach.

QUICK HITS

💸 OpenAI Hires Uber India's Former Chief to Run Its Second-Biggest Market.

☁️ The Compute Crunch Hits Big Tech as Google Rations Meta's Gemini Use.

🍎 OpenAI Poaches Apple's Vision Pro Boss to Build Its AI Devices.

🔧 Qualcomm Eyes $10B for Tenstorrent, Tripling the Chip Startup's Value in a Year.

 *partner link