Claude Opus 5.5 takes first place on the Artificial Analysis Intelligence Index

Anthropic's Claude Opus 5.5 leads an independent intelligence ranking, and Google shows its Genie 4 world model. Apple brings an AI assistant to its office apps, while compact OCR and audio models arrive from ATH-MaaS, OpenBMB and Liquid AI.

1. Claude Opus 5.5 takes first on the Artificial Analysis Intelligence Index at 58

   Anthropic's new model now leads an index scored at 58 by Artificial Analysis, which runs its own evaluations instead of relying on lab reports.

   Models: Claude Opus 5.5

2. Google unveils Genie 4 world model for interactive experiences

   Google's world model adds audio, more realistic worlds and real-time prompt updates, with experiences lasting about two minutes instead of one.

3. Apple adds an AI assistant to Pages, Keynote and Numbers

   Built on Apple and OpenAI models, the assistant edits slides and builds charts, and Model Context Protocol support lets third-party agents connect on macOS 27.

4. Stanford professors say their books were pirated to train Claude

   A Stanford Daily report says Anthropic trained Claude on illegally obtained books by Stanford professors, a claim that remains a report, not a confirmed finding.

5. OvisOCR3, a 0.8B OCR model, reports top scores on several document benchmarks

   At just 0.8B parameters, OvisOCR3 handles parsing, OCR, extraction and table reconstruction in one model and scores 96.74 on OmniDocBench v1.6.

6. OpenBMB releases MiniCPM-OCR 1.0 and MiniCPM-OCR Jev open-source models

   OpenBMB pairs a 1.3B OCR model scoring 93.68 on Real5-OmniDocBench with a 2B quality judge, both supported by vLLM, SGLang, llama.cpp and Ollama.

7. Liquid AI releases LFM3-Audio-2B, a compact audio model for local devices

   The 2B model runs on laptops and edge devices, offering speech recognition, translation across 18 languages and voice-activated tool calling.

8. Hugging Face publishes LFM3 for building specialized multimodal models

   Liquid AI positions LFM3 as a family for building specialized multimodal models, according to a Hugging Face blog post.

Written by Claude (claude-opus-5-5) from the stories below.