At Research@ 2026, Google opens Deep Review to the public: an agentic checker for papers and technical documents, piloted on 16,000+ conference submissions; Earth AI agent API enters early access

Why it matters

【Key points】

At Research@ 2026 in Mountain View on Oct 14, Google Research publicly launched Deep Review, an agentic system for checking complex documents. It grew out of the Paper Assistant Tool, which gave opt-in authors pre-submission feedback on 16,000+ manuscripts at STOC, ICML, NeurIPS and ICLR, separate from official review. It segments a document, checks logic and math, grounds facts and compares with prior work. Now on Gemini 4 Argon, it also covers legal and financial documents. Two free reviews a day; more via the user's own Gemini API key and a Deep Review API. Google says uploads are never used for training.

Also announced: an Earth AI geospatial agent technical report, two new benchmarks and an early-access agent-to-agent API; TimesFM and TabFM coming to Gemini for business; a Synaptics edge AI chip on Google's open Coral NPU IP, due next week.

【Verifiable】

Deep Review: https://research.google/blog/analyzing-technical-documents-with-deep-review/

Research@ top 10: https://research.google/blog/research-2026-10-ways-were-amplifying-human-ingenuity/

Earth AI agent: https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/

【Our read】

1. A real bottleneck: by Google's count ICLR, NeurIPS and ICML drew 74,000+ submissions this year versus about 23,000 in 2023, while reviewer capacity is flat. A public checker will first change how authors self-check, not replace reviewers.

2. Limits: it checks logic, math and grounding; novelty and value stay with humans. Pilot feedback is Google's own, with no independent error rates yet, and no conference has said whether it allows or requires its use.

3. Free daily quota plus billing to the user's Gemini key turns a research prototype into an API on-ramp; Earth AI's A2A API follows the same path.