AI Tech Daily March 30 2026: PaddleOCR tops GitHub, New Large Models, and Security Advances

1. PaddleOCR tops GitHub OCR rankings – Baidu’s PaddleOCR amassed over 73.3 K stars on GitHub, overtaking Google’s Tesseract OCR. The project leverages deep‑learning techniques, delivering strong performance and expanding market share, which the author notes as a significant Chinese breakthrough in optical character recognition.

2. GigaWorld‑1 model leads global contests – GigaAI released the full‑stack, self‑developed embodied foundation model GigaBrain‑0.1, which topped the RoboChallenge leaderboard. Its next‑generation native paradigm GigaBrain‑0.5M demonstrated self‑evolution and superior performance on long‑duration tasks, indicating a major advance for embodied AI in China.

3. Eli Lilly invests 2.75 billion RMB in AI drug discovery – The pharmaceutical giant pledged a 2.75 billion RMB stake in UK‑based AI startup Yingxi Intelligence, aiming to accelerate drug‑development pipelines. The author emphasizes the investment as evidence of AI’s growing influence in pharmaceutical R&D.

4. Doubao joins the top‑tier LLM cohort – ByteDance’s Doubao model entered the global first‑tier large‑model lineup, with Xiaomi’s MiMo appearing for the first time. Chinese models showed marked improvements in Chinese language understanding, drawing industry attention toward parity with GPT‑4.

5. InfoTok redefines video tokenization – Researchers from Stanford and other institutions introduced InfoTok, a video‑segmentation technique that uses dynamic compression rates to drastically cut computational redundancy while increasing token information density, a development the author expects to shape future video‑understanding and generation tasks.

6. Claude cracks a 20‑year‑old vulnerability in 90 minutes – The AI company Claude demonstrated the ability to exploit a two‑decade‑old security flaw within 90 minutes, showcasing unprecedented breach speed. The author notes the system’s multi‑million‑dollar funding and predicts significant implications for cybersecurity defenses.

7. New Claude model outperforms Opus 4.6 – A mysterious upcoming Claude model was revealed, with the author stating it surpasses the performance of Opus 4.6 across all measured metrics.

8. Node.js core members petition to ban AI‑assisted development – Over a hundred core contributors signed a petition demanding the prohibition of AI tools like Claude Code (1.9 k lines) within the Node.js project, raising concerns about software‑development ethics and future AI governance.

9. Kimi K2.5 model exceeds $1 billion monthly revenue – The “Moon’s Dark Side” Kimi K2.5 model generated over $1 billion in a single month, driven by its “thinking‑type” capabilities that attract enterprise customers willing to pre‑pay tens of millions for compute, highlighting AI’s commercial potential.

10. Tiny AI model achieves 15× inference speed on a single GPU – An AI startup released a compact model that delivers a 15‑fold inference speed increase on one graphics card while maintaining 93.8 % accuracy, surpassing DeepSeek‑V3.2 and markedly reducing R&D costs for enterprises.

11. Claude’s 15 KB engine boosts rendering speed 1,200× – Claude introduced a 15 KB rendering engine that accelerates rendering by a factor of 1,200, dramatically lowering resource consumption and promising a new era of lightweight engines for mobile game development.