Monthly report

August 2026

48.1 / 100

+2.2 from July

Confidence: Medium

August continued the upward trend, led by coding and agents, while stronger evaluation practices added a modest reliability signal.

Category Scores

  • Agents53
  • Coding60
  • Reasoning47
  • Science35
  • Multimodal49
  • Robotics27
  • Reliability38

Top signals this month

  1. Long-horizon agentic workflows became more concrete.

    xAI released Grok 4.6 for coding, agentic tasks, and knowledge work, alongside Grok Bot for persistent cloud-based agent workflows.

  2. Gemini 3.7 Flash improved coding performance.

    Google reported sizable gains over Gemini 3.6 Flash on debugging, issue resolution, and production-code evaluations.

  3. Science and evaluation quality both moved forward.

    WeatherNext reported state-of-the-art cyclone forecasting results, while DeepMind piloted double-blind frontier-model evaluations designed to reduce benchmark contamination.