July 30, 2026

AI News Brief

Daily AI news, sourced from Brave News API
← All Briefs
01

Moonshot open-sources Kimi K3, a 2.8-trillion-parameter model billed as the first open "3T-class" AI

Moonshot AI has open-sourced Kimi K3, a 2.8T-parameter model it calls the first open "3T-class" AI. It posts 88.3% on Terminal-Bench 2.1 (half a point behind GPT-5.6 Sol, ahead of Claude Fable 5 and Opus 4.8) and trades blows with frontier closed models Claude Fable 5 and GPT-5.6.

02

Zuckerberg urges US not to ban Chinese AI models; 20+ companies (Meta, Nvidia, Microsoft, Palantir) push to protect open-weight AI

Meta CEO Mark Zuckerberg said banning Chinese AI models is "not an effective solution" and warned against "regulatory capture," calling for fewer barriers. A coalition of 20+ tech companies released a joint letter urging policymakers to avoid restrictions on open-weight AI models, arguing openness expands access to the AI economy.

03

Nvidia backs Reflection AI with $800M to lead Western open-source AI

Nvidia led an $800M round in Reflection AI, positioning it as a champion of Western open-source AI in direct response to China's rapid advances (DeepSeek, Kimi K3). The bet underscores that open-source AI is now a strategic front in the US-China AI race.

04

Samaya AI launches FrontierFinance benchmark; open-source GLM 5.2 and DeepSeek evaluated against frontier

Samaya AI released FrontierFinance, a new public benchmark for AI agents in investment workflows. Samaya's own system outperformed Claude Fable 5 and GPT-5.6 Sol; the evaluation included open-source models GLM 5.2 and DeepSeek alongside frontier closed models from Anthropic, OpenAI, and Google.

05

China's cheap AI models (Kimi K3, Qwen3.8 Max) close the gap on ChatGPT and Claude at a fraction of the cost

Vox reports Chinese companies released two models - Kimi K3 and Qwen3.8 Max - that are nearly as powerful as ChatGPT and Claude but much less expensive, pressuring OpenAI and Anthropic's closed-loop business model. GLM-5.2 already offers a one-million-token context and open weights at lower economics.

06

Amazon winds down most Nova AI models, redirects talent to one frontier effort

Amazon is phasing out most of its Nova AI models and redirecting talent to a single frontier effort, signaling that model quality now matters more than portfolio size in the AI race.

07

Samsung lands $200B Broadcom AI chip deal; Q2 profit beats estimates on soaring AI chip demand

Samsung secured a Broadcom AI chip manufacturing deal worth more than $200B through 2030, and reported Q2 operating profit that beat estimates on soaring AI server demand, with all-time high DRAM and NAND sales. Samsung also warned chip supply shortages will worsen due to the AI boom.