Kimi K3 matches frontier models as White House Gold Eagle plan draws fire

New CNBC reporting details the Trump administration's Project Gold Eagle framework for controlling access to frontier AI models, drawing sharp criticism from former White House AI czar David Sacks who warned the policy threatens US leadership. Chinese startup Moonshot AI released full specs and pricing for its Kimi K3 model, claiming parity with Anthropic's Fable 5 on key benchmarks, while Google delayed Gemini 3.5 Pro but unveiled its record-setting Aletheia AI mathematician. Anthropic capped Fable 5 usage for paid subscribers and merged its Chat and Cowork interfaces, as a Bloomberg report revealed OpenAI is developing an always-on AI speaker device.

Written by AI from the sources it cites; it can be wrong — follow any citation to confirm.

13 sources2 min read

Gold Eagle framework tightens White House grip on frontier AI model access The Trump administration formalized Project Gold Eagle as a government clearinghouse requiring explicit approval for partner access to frontier models from Anthropic and OpenAI, following last month's block of Claude Mythos 5. Former AI Czar David Sacks warned the move could cede the AI race to China, calling it a self-imposed slowdown just as Chinese labs accelerate. 1 2 3

Moonshot's Kimi K3 debuts as largest open-weight model, challenges US labs The 2.8-trillion-parameter multimodal model provides a 1-million-token context window and open weights on July 27, with API pricing ($3/$15 per million tokens) drastically undercutting US rivals. Moonshot claims K3 tops independent coding benchmarks and matches or exceeds Claude Fable 5 on several internal evaluations. 4 5 6 7

Anthropic limits Fable 5 access while merging Chat and Cowork modes Starting July 20, Fable 5 is capped at 50 percent weekly usage for Max and Team Premium subscribers, while Pro users lose included access in favor of a one-time $100 credit. Separately, Anthropic unified its conversational Chat and autonomous Cowork modes into a single interface on web and desktop. 8 9 10

Google delays Gemini 3.5 Pro but launches Deep Think update and Aletheia Internal sources confirmed Gemini 3.5 Pro's launch slipped due to insufficient coding performance, a blow to Google's flagship roadmap. Simultaneously, Google updated Gemini Deep Think to outperform GPT-5.2 on science benchmarks, and DeepMind released Aletheia, an AI agent that solved 91.9 percent of Olympiad-level math proofs. 11 12

OpenAI develops screenless always-on AI speaker for 2027 launch Bloomberg reports OpenAI is building a portable, screenless speaker equipped with cameras and sensors designed to anticipate user needs proactively rather than waiting for prompts. The device represents a major hardware push but faces legal uncertainty from Apple's ongoing trade secrets lawsuit against the company. 13