DeepSeek launched a preview of a new model adapted for Huawei chip technology, signaling China's push for semiconductor autonomy in AI. The move underscores the strategic shift toward domestic hardware integration following last year's viral low-cost model success.

Ineffable Intelligence closed a $1.1B seed round co-led by Sequoia and Lightspeed, with backing from Nvidia, Google, and the U.K.'s Sovereign AI Fund. The record seed size reflects investor appetite for reinforcement learning-focused startups targeting AGI research.

Anthropic opened its Sydney office and appointed Theo Hourmouzis as GM for Australia & New Zealand. The expansion marks Anthropic's geographic scaling beyond North America and Europe as Claude adoption grows in the region.

Alibaba ATH Innovation Unit began grayscale testing of HappyHorse 1.0 (快乐小马), a video generation and editing model, on April 27. The release signals continued competition in the multimodal space with Chinese alternatives to OpenAI's Sora and other Western video models.

OpenAI secured a $20B capacity purchase agreement spanning 2026–2028, with option for an additional 1.25 gigawatts by 2030, signaling massive scaling of production inference workloads. The deal demonstrates how frontier labs are locking in long-term compute commitments to meet deployment demand.

China ordered Meta to divest its acquisition of Manus, a Singapore-based AI startup with Chinese founders, effective immediately on April 27. The move signals Beijing's tightening control over foreign investment in domestic AI and may deter Chinese entrepreneurs from seeking overseas partnerships.

DeepSeek released a preview of its next-generation model adapted for Huawei chip architecture on April 24, demonstrating China's push toward semiconductor independence. The release positions DeepSeek to challenge OpenAI, Anthropic, and Google while circumventing U.S. chip restrictions.

Anthropic made Claude Mythos Preview available to 11 selected organizations on April 7 for identifying and fixing cybersecurity vulnerabilities. The limited preview gates access to the next-generation model and focuses deployment on high-value enterprise use cases.

Tencent launched Hy3, an open-source mixture-of-experts model, in late April after less than three months from training cold start—unusually rapid for a frontier model. The efficiency breakthrough and accelerated timeline suggest competitive pressure in the Chinese LLM market intensifying.

OpenAI released GPT-5.5, an agentic model designed to autonomously work through complex tasks by switching between multiple tools, now available via API as of April 24. The model is positioned as OpenAI's smartest and most intuitive offering to date, with improved coding and research capabilities.

DeepSeek released DeepSeek-V4, a new open-source Mixture-of-Experts model series claimed as the most powerful open-source platform, challenging offerings from OpenAI and Anthropic. The release marks a significant shift in open-weight model capability, with DeepSeek matching frontier proprietary models on multiple benchmarks.

OpenAI's official announcement confirms GPT-5.5 and GPT-5.5 Pro are now available in the API as of April 24, 2026, with updated system cards describing additional safeguards. GPT-5.5 represents a new class of agentic intelligence with enhanced autonomy for multi-tool task execution.

SpaceX announced an option to acquire Cursor, maker of AI-powered coding software, for $60 billion in common stock and equity awards, preempting a planned $2 billion funding round. The deal signals major tech conglomerates consolidating AI developer tools and infrastructure.

Google announced a $40 billion investment in Anthropic, starting with $10 billion initial funding and $30 billion in future commitments over multiple years. The mega-round reflects intensifying competition for frontier AI capabilities among Big Tech players.

OpenAI released GPT-5.5, an agentic model designed to autonomously work through complex tasks by switching between multiple tools, with added guardrails restricting cybersecurity applications. Developers need to understand the new capability tier and usage restrictions for production deployments.

Anthropic announced Claude Mythos Preview as part of Project Glasswing, committing $100M in model usage credits for security research, with pricing at $25/$125 per million tokens for preview participants. Researchers and security teams gain access to a frontier model explicitly designed for critical infrastructure auditing.

OpenAI released an open-weight Privacy Filter model for detecting and redacting personally identifiable information in text as infrastructure for the software ecosystem. Developers building compliance-critical applications can now integrate tested PII detection without proprietary dependencies.

Tencent released Hy3, an open-source frontier-class MoE model, in under three months from cold start training in late January 2026. The unusually fast development cycle demonstrates accelerating timelines for competitive open-weight LLM releases.

SpaceX secured an option to acquire AI coding software maker Cursor for $60 billion or partner for $10 billion, preempting a $2 billion funding round. The consolidation signals major capital allocation toward agentic AI coding tools ahead of SpaceX's planned IPO.