Daily Research Brief 2026-07-12

📊 Token usage: estimated from retrieval and writing scale.

Covers the latest AI research, open source and industry moves, updated daily.

Editor’s Note

Two signals worth practitioners’ attention today. First, ‘self-evolution’ is moving from narrative to documented fact: GPT-5.6’s lightweight version Luna was trained entirely autonomously by the flagship Sol (finding GPUs, setting configs, writing scripts, verifying — no humans in the loop), combined with open-source self-evolving agents like NousResearch/hermes-agent — the human structure of model iteration is being rewritten; small and mid teams should care more about ‘how to orchestrate’ than ‘how to train’. Second…

1. Latest arXiv Papers

1. Cognitive-structured Multimodal Agent for Multimodal Understanding, Generation, and Editing — https://arxiv.org/abs/2607.08497

2. PairCoder++: Pair Programming as a Universal Paradigm for Verified Code-Driven Multimodal and Structured-Artifact Generation — https://arxiv.org/abs/2607.01883

3. Seeing and Reflecting: Multimodal Memory-Enhanced Agent Collaboration for Recommendation — https://arxiv.org/abs/2607.07108

4. Perceive-to-Reason: Decoupling Perception and Reasoning for Fine-Grained Visual Reasoning — https://arxiv.org/abs/2607.01191

5. HIVE: Understanding Post-Hallucination Reasoning in Vision Language Models — https://arxiv.org/abs/2607.07507

6. DeltaV: Thinking with Visual State Updates in Unified Large Multimodal Models — https://arxiv.org/abs/2607.08434

7. ProLaViT: Learning Progressive Latent Visual Thoughts in Structured Latent Space — https://arxiv.org/abs/2607.02907

8. Multimodal Continuous Reasoning via Asymmetric Mutual Variational Learning (AMVL) — https://arxiv.org/abs/2607.00461