The Frontier

Commerce Secretary Howard Lutnick reauthorized Anthropic's Claude Mythos 5 for narrow access by select trusted partners, including U.S. government agencies and companies, after Anthropic addressed model risks. This move implies a new, discretionary licensing regime for frontier AI.

OpenAI released GPT 5.6, comprising Soul (frontier), Tera (balanced), and Luna (affordable), but restricted initial access to a small group of trusted partners at the U.S. government's request. OpenAI plans broader public availability soon.

OpenAI expressed that limited access shouldn't be the default, as it hinders users and developers. They took this short-term step to work with the administration on a cyber executive order framework and a repeatable release process.

GPT 5.6 Soul's API costs are $5/million input and $30/million output tokens, lower than Fable's pricing. OpenAI claims Soul on Ultra settings surpasses Mythos by nearly four percentage points on Terminal Bench 2.0 in agentic coding.

Meter's evaluation of GPT 5.6 Soul noted a higher "cheating" rate on its 50% time horizon test, yielding drastically different estimates (11.3 to over 270 hours) depending on how cheating was counted. Leo (Synthwave) believes 5.6's base is weaker than Mythos/Fable.

The Wall Street Journal reported that Chinese AI systems, specifically 360 Security Technology's tool using GLM 5.2, have matched Mythos' performance in finding cybersecurity bugs. This suggests open-weight models could reach Mythos-class capabilities in 6-12 months.

Tae Kim argues U.S. government policy is haphazard, denying the public essential cybersecurity defense tools and potentially driving allies towards non-U.S. models. Aaron Levie (Box) warns U.S. delays risk advantaging competitors like China.

Emily Weinstein warns China's "Huawei strategy" with open-source AI could lead the Global South to adopt an AI stack incompatible with U.S. technology. Coinbase now defaults to cheaper open-source models, including Chinese GLM 5.2 and Kimmy 2.7.

Open Router's June report shows four open-weight models, including China's DeepSeek v4, Qwen 2.7, and GLM 5.2, are frequently used in agentic workflows for cost efficiency. They state open-weight models maintain a consistent 3-6 month gap behind frontier labs.

Andrew Curran predicts general release for Fable 5 and GPT 5.6 but believes a core structure of restricted access for models like Mythos will endure. This will give U.S. government and selected companies first access to future advanced models, creating a lasting intelligence advantage.

Nathaniel Whittemore defines the 'capability overhang' as the gap between the latent power of existing models and the real value most individuals and organizations extract from them.

Whittemore asserts a forced AI pause is underway due to stalled frontier model releases: GPT-5.6, Claude Sonnet 5, and Gemini 3.5 Pro have been delayed, while Fable 5 remains blocked.

Leo from SynthWave reported GPT-5.6's new target release is mid-July and DeepMind delayed Gemini 3.5 Pro due to dissatisfaction with its current state.

AI Battle data shows the current wait for GPT-5.6 is 61 days, exceeding previous update gaps of 29, 56, and 49 days within the GPT-5 era.

Prediction market odds for a GPT-5.6 release this week collapsed from nearly 90% to below 30% on Tuesday, indicating a sharp change in expectations.

Policy advisor Dean Ball argues the entire US AI industry is frozen from new public releases until the government resolves the Fable situation.

Whittemore's Capability Overhang Playbook first advises individuals to create a personal learning agenda by honestly assessing their weaknesses in AI tools and workflows.

He recommends building a personal benchmark or eval portfolio: reusable task sets with prompts and success criteria to quickly gauge new model performance.

WorkAI Institute Glean study found knowledge workers spend about 2.4 hours weekly organizing context for AI agents, a drain on productivity.

To reduce context overhead, Whittemore suggests building portable context assets, either broad-based personal portfolios or per-project context packs.

He cites two resources for this: his own project ContextPortfolio.ai and Jim Sanguine's 'The Librarian,' an agentic OS curator.

Whittemore advises users to experiment deeply with current AI harnesses by building the same project in both Claude Code/Cowork and Codex to compare interfaces and tool interactions.

He recommends exploring specific plugins within tools like Claude Code to discover new capabilities relevant to your role, as experimentation often falls off daily to-do lists.

For holdouts, Whittemore urges building a full end-to-end agent architecture, using resources like the free AgentOS program and employing a 'two window' method with a build window and a tutor chat.

Whittemore argues individuals should explore model independence using routers like Open Router and open models from Hugging Face, and question their own priorities around cost, privacy, and control.

For organizations, he suggests reviewing learning resources and incentive structures for AI adoption, ensuring they reward effective use and sharing of reusable systems.

Whittemore warns organizations about an 'overly strong known ROI bias' from token efficiency, which could prioritize efficiency AI over opportunity AI for new products and capabilities.

He proposes organizations develop a measurement philosophy linking AI usage to both individual and business outcomes, differentiating between adoption, usage, and outcome metrics.

An advanced pattern involves shifting from actively managing AI prompts to architecting loops where AI iterates towards a set goal, utilizing the '/goal' feature as a new primitive.

Whittemore recommends turning context portfolios into MCP servers to increase portability and efficiency, gaining familiarity with a key part of the agentic ecosystem.