AI News Brief

[ROCKET] Moonshot AI Releases Kimi K3: World's Largest Open-Weight Model at 2.8T Parameters

Chinese startup Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter open-weight model with a 1-million-token context window - the largest open AI model to date. Moonshot claims it delivers competitive performance against Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, and "substantially outperformed" Claude Opus 4.8 and GPT-5.5; full weights are scheduled for release by July 27.

[US_FLAG] Mira Murati's Thinking Machines Lab Ships Inkling, the Largest US Open-Weight Model

Former OpenAI CTO Mira Murati's Thinking Machines Lab released Inkling, a 975-billion-parameter Apache 2.0 multimodal model supporting text, image, and audio inputs with a 1-million-token context window. It scored 97.1% on the AIME 2026 benchmark and is positioned as a US-developed open-weight alternative built for enterprise fine-tuning rather than benchmark-chasing.

[CHART_UP] GLM-5.2 Captures 40% of Developer Tokens as Open Weights Reshape the Market

Z.ai's GLM-5.2 now accounts for roughly 40% of developer tokens routed through OpenRouter, scoring 74.4 on FrontierSWE (within one point of the closed-model frontier) and 54.7 on Humanity's Last Exam - ahead of GPT-5.5's 52.2. The model reportedly costs around one-sixth what GPT-5.5 charges per task, with a GLM-5.5 release rumored for August.

[GLOBE] Xi Jinping Pushes Open-Source AI, Launches WAICO Alliance at World AI Conference

At the World Artificial Intelligence Conference in Shanghai on July 17, Chinese President Xi Jinping called for a "principle of openness" in AI development and announced the WAICO (World AI Cooperation Organization) alliance. Beijing is expected to use WAICO to influence global AI regulation, framing China's open-weight releases as a counter to US closed-model dominance.

[CAPITOL] White House Takes Control of Frontier AI Model Access from OpenAI and Anthropic

The Trump administration is now dictating who can access frontier AI models, taking control previously held by OpenAI (Project Daybreak) and Anthropic (Project Glasswing) under their cyber initiatives. The move comes as AI czar David Sacks champions a "hands-off" regulatory approach, warning the US could lose the AI race after Kimi K3 became the first Chinese model to top the Frontend Code Arena.

[CHART_DOWN] AI Chip Stocks Slide as Kimi K3 Triggers "DeepSeek Shock 2.0" Selloff

Following Kimi K3's release, AI chip stocks tumbled: SoftBank dropped over 9% and the Nasdaq fell 1.47% as semiconductor shares came under pressure. The selloff renewed concerns about ROI on AI infrastructure build-outs, with some analysts arguing cheaper open models could trigger a spending pullback on chips. Separately, Apple is reportedly in talks to acquire AI chip startups to bolster its server infrastructure after struggling to run Gemini models on its own silicon.