Today’s frontier report
Research Pool → Multi-Scout → Editor → Edition
3qk7 · AI LOCAL LABS TECHNICAL SCOUTING

AI FRONTIER

Signals worth understanding before they become obvious.

2026-11-09
01
testinferencecost collapse

OpenRouter new model: Auto Router

Auto is OpenRouter's flagship model router, built for general-purpose usage, adapting to each request by choosing cost-optimal models and adjusting reasoning effort while balancing task difficulty, benchmark results, live prices, and prompt caching.

  • Auto Router has no fixed price of its own; each request is billed at the price of the model that serves it.
  • Auto Router can route to models with context windows of up to 2,000,000 tokens, and the context length available to a request is that of the model that serves it.
  • It accepts text, images, audio, files such as PDFs, and video as input and returns text and images.
  • Auto Router was released on November 5, 2026, and is not currently available through EU in-region routing.

For builders, Auto shifts model selection and reasoning-effort tuning to the router, so cost varies with the routed model and budget control depends on governance settings rather than a fixed unit price.

OpenRouter Models
02
watchagentscapability unlock

Google strengthens CodeMender, its vulnerability-fixing AI agent built on Gemini 4 Argon

Google has strengthened CodeMender, its AI agent for fixing vulnerabilities, which uses Gemini 4 Argon.

  • CodeMender is Google's AI agent for fixing vulnerabilities.
  • CodeMender uses Gemini 4 Argon.

For builders, this shows Google pairing Gemini 4 Argon with an automated vulnerability-fixing agent, a reference point for agent design.

ビジネス+IT
03
watchresearchcapability unlock

Anthropic AI reportedly solved a quantum problem open since 2008

According to a report citing a paper abstract, an artificial intelligence from Anthropic reportedly solved a quantum problem that had remained open since 2008.

  • The source is a paper abstract reported by Opinando San Nicolás.
  • The problem is described as open since 2008.

If confirmed, this would show an AI system tackling a long-standing open quantum problem, making it worth watching whether the method generalizes to other unsolved problems.

Opinando San Nicolás
04
watchresearchincremental

DeepMind discloses two flaws in RSA 1024-bit, RSA 2048-bit remains secure

According to the paper abstract, DeepMind disclosed two flaws in RSA 1024-bit while stating that RSA 2048-bit remains secure.

  • The paper abstract states that DeepMind disclosed two flaws in RSA 1024-bit.
  • The paper abstract states that RSA 2048-bit remains secure.

For AI builders, this signal suggests systems still relying on RSA 1024-bit warrant migration review, while RSA 2048-bit is not listed as affected in this abstract.

Pasquale Pillitteri
05
watchresearchincremental

Detecting signs of hypertension using wearable sensors

Google Research presents AI-PPG-ACC-HTN, a multimodal AI system that uses PPG and accelerometer signals from a consumer smartwatch to passively detect a persistent hypertensive state without cuff calibration, validated in a Nature paper against 24-hour ambulatory blood pressure monitoring.

  • The waveform encoder was pretrained with self-supervised contrastive learning on 492,347 hours of unlabeled sensor data from 2,109 consenting participants, using 15-second PPG and ACC waveforms.
  • For initial screening the smartwatch achieved 65.8% sensitivity, 90.0% specificity and 80.6% PPV, versus 55.3% sensitivity and 77.8% PPV for initial office-based screening.
  • In the confirmatory subset (N=48) the system achieved 78.4% sensitivity, 90.9% specificity and 96.7% PPV, versus 67.6%, 63.6% and 86.2% for return office measurements.

For AI builders, the work shows a template of pretraining a waveform encoder on unlabeled wearable data and aggregating multi-day person-level embeddings for classification, yielding an individual risk score without cuff calibration.

Google Research