NEW RUNTIME
SignalNov 5, 20261 min read4 sources

Low reasoning effort should be the default, with escalation only when a task fails

In one minute

Vals AI found lower reasoning settings often cut run cost far more than they cut scores, and in 22 of 56 cases matched the leaderboard result.

A separate study found the highest setting cost up to 42 times more on easy tasks and still lost some tasks that lower effort solved. LoCo trims reasoning output by roughly 40% without losing accuracy, and Claude Code now sets effort per fork.

From the Field Notes:

EditorAndrey Reshetnikov, AI product and AI engineering

  1. Nov 10, 2026Claude Fable 5.2 scored 174 on the Epoch Capabilities Index, beating GPT-6 Astra's prior record of 169.claudegemini
  2. Nov 7, 2026Artificial Analysis launched VIALS-AA, a leaderboard for scientific image reasoning across 161 tasks and 8 domains.claudegemini
  3. Nov 6, 2026Code released as a HELM PR and standalone dspy-helm pipeline.claudegemini
  1. Agent economics moves to completed work

Evidence / sources

  1. 1vals.ai/blogs/reasoning-effort
  2. 2arxiv.org/pdf/2610.34491
  3. 3arxiv.org/abs/2611.04814
  4. 4github.com/anthropics/claude-code/releases/tag/v2.1.318
All Signals

Next reads

Read next

Open the graph
  1. 01topicClaudeExplore the claude topic hub.
  2. 02topicGeminiExplore the gemini topic hub.
  3. 03archiveSignalsOpen the signals and the raw-signal archive.
  4. 04sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract