Washington, Beijing Gate AI Models.

Mazen Al Sabbagh

— 4 min read
Share

TL;DR

  • GPT-5.6 Cleared: Washington lifted its national security hold on OpenAI’s next flagship model; broad release is set for Thursday after weeks of restricted access.
  • Beijing's Mirror Move: China is considering export controls on its advanced AI models from Alibaba, ByteDance, and Z.ai, mirroring US strategic restrictions.
  • Chinese Models Surge Stateside: Chinese AI models now command a significant share of US enterprise token traffic on OpenRouter, prompting a Congressional probe.
  • GPT-Live Ships: OpenAI’s full-duplex voice models roll out in ChatGPT, enabling simultaneous listening and speaking.
  • Blue Origin Raises: The first external funding round for Bezos's venture secures $10B at a $130B valuation, with Coatue leading.

Lead Story: Washington, Beijing Gate AI Models

OpenAI's GPT-5.6 will see a broad release this Thursday, following the US government's cessation of limits on the model's deployment. This action concludes a security review prompted by concerns over potential misuse in cyber warfare and biotechnology. Previously, access was restricted to approximately 20 vetted organizations and government evaluators. The clearance establishes GPT-5.6 as the first US frontier model to navigate a pre-release government gating process before general availability, shipping in Sol, Terra, and Luna versions.

Beijing converges on a similar strategic posture from the opposite direction. Chinese officials are reportedly discussing restrictions on overseas access to advanced models from domestic firms like Alibaba, ByteDance, and Z.ai, including future open-weight releases. This initiative directly mirrors US export control mechanisms.

The timing of these state interventions is not coincidental. A recent CNBC investigation revealed that Chinese models have captured 30–46% of US enterprise token volume on OpenRouter weekly since February, a substantial increase from 4.5% in the first half of 2025. DeepSeek alone routes 5.13 trillion tokens each week, establishing itself as the platform's largest single vendor, with pricing as low as $0.14 per million input tokens against GPT-5.5's $5.00. Today, US lawmakers initiated a probe into this dependence. Both Washington and Beijing now classify model weights as strategic goods, dictating their cross-border movement on state terms.

In Other News

OpenAI Debuts GPT-Live. ChatGPT's voice interface now incorporates full-duplex models, enabling simultaneous listening and speech generation. This upgrade eliminates the former walkie-talkie cadence, allowing for natural interruptions and back-channel communication. GPT-Live-1 is the default for Go, Plus, and Pro subscribers, with a mini version for free users and GPT-5.5 handling more complex processing. An API version is confirmed, specifically targeting the modality where voice assistants have historically felt underdeveloped.

Meta's Superintelligence Labs Ships Muse Image. Meta AI introduced its first image model, Muse Image, designed as an agent that integrates search and coding tools, self-refines outputs, and scales quality through test-time compute. It has rolled out across the Meta AI app, Instagram Stories in the US, and WhatsApp in select countries, featuring invisible Content Seal watermarking. A preview of Muse Video was also offered for future release. User pushback concerning photo data utilization has already emerged.

Blue Origin Secures First External Funding. Jeff Bezos’s space company closed its initial external funding round, raising $10 billion at a $130 billion pre-money valuation. Coatue led the round with approximately $4 billion, complemented by a $2 billion investment from Bezos himself. This capital injection follows a New Glenn rocket explosion during testing six weeks prior, with Bezos and CEO Dave Limp targeting a return to flight by year-end.

Geneva Concludes All-Nations AI Summit. The UN's inaugural Global Dialogue on AI Governance wrapped its session with the Scientific Panel's Yoshua Bengio cautioning that advanced AI poses a risk of "catastrophic harm" that science cannot fully guarantee against. Secretary-General Guterres pressed for binding directives on autonomous weapons. The AI for Good Summit continues through Friday in the same venues, with key industry figures present. The next session is scheduled for New York in May 2027.

X / Social Pulse

Sam Altman articulated the significance of GPT-Live day, noting, "it feels magical and 'real'. i have always preferred typing to talking to an AI, now i think that's going to shift." Early reactions, surpassing 500K views, were largely positive, though a dissenting thread is developing around the model's perceived "over-enthusiasm"—users report it gushing over routine queries. The other active debate centers on whether Thursday's government-sanctioned GPT-5.6 launch constitutes a genuine security triumph or a strategic market barrier disguised as one.

One to Watch

GPT-5.6 Market Impact. Thursday's GPT-5.6 release will provide crucial pricing and benchmark details for its Sol, Terra, and Luna tiers. The primary observation point is whether this government pre-clearance establishes a new template for all frontier labs, effectively initiating a de facto licensing regime without explicit legislation. Concurrently, Google's Gemini 3.5 Pro remains in enterprise preview with no confirmed general availability date, having slipped from its original June target.

Quick Hits

  • Apple Loses EU DMA Challenge: The European General Court rejected Apple's privacy and security arguments, upholding its designation as a gatekeeper under the Digital Markets Act.
  • Microsoft Shifts to In-house AI: Microsoft is now routing tens of thousands of weekly Excel and Outlook prompts to its internal MAI models, displacing OpenAI and Anthropic due to cost efficiencies.
  • SambaNova Raises $1 Billion: The AI inference acceleration specialist secured $1 billion at an $11 billion valuation, led by General Atlantic.
  • Temasek Lifts AI Exposure: Singapore's state fund plans to increase its AI exposure from 6% to 15% of its record S$518 billion portfolio over the next five years.
  • ESRB Warns on AI Cyber Risk: The European Systemic Risk Board cautioned that frontier AI could substantially amplify cyber risks for financial institutions, accelerating vulnerability discovery, phishing, and malware deployment.

Frontier AI spent years as a product category; this week, it formally became a customs category. When sovereign approvals dictate release schedules and export intentions shape market access, the autonomy of AI development is fundamentally recalibrated.

Sources

GPT-5.6 & US-China model politics: CNBC (GPT-5.6 clearance) · CNBC (Chinese model adoption) · CNBC (lawmaker probe) · Tech Startups roundup GPT-Live: OpenAI · TechCrunch · VentureBeat · @sama · BigGo Meta Muse: Meta Newsroom · Meta AI blog · CNBC · TechCrunch Blue Origin: CNBC · TechCrunch · Bloomberg Geneva: UN News (Dialogue) · UN News (Guterres) · ITU Other: Build Fast with AI

Lock in. M. mazen@thorterminal.com

Read more