Claude Code: Model Does Not Equal Effort
新智元 · wechat · 2026-07-12
This article explains Anthropic's clarification regarding Claude Code: Model and Effort are not the same.
- Model determines the "brain"—the capability ceiling of the model itself; changing models essentially means changing weights.
- Effort determines the "input"—how many files the model is willing to read for a given task, whether it runs tests, if it repeatedly verifies, and how thoroughly it executes long task chains.
The article reviews the March incident where users complained about Claude Code "getting dumber": the default Effort was downgraded from high to medium, leading many to mistakenly believe the model had degraded, though it was actually a shift in behavior tiers. Anthropic later reverted the default and reset quotas in April.
Core conclusions:
- First, check if sufficient context is provided and tools are properly configured.
- If the model "doesn't know how," switch to a more capable model.
- If it "isn't trying hard enough," prioritize increasing Effort.
The article also emphasizes that a smaller model + high Effort can outperform a larger model + low Effort on many tasks. Meanwhile, the AI programming competition is shifting from "comparing model leaderboards" to "comparing orchestration capabilities" and "how work is delegated to agents."
More from coding & agent
- Multiagent v2 playbook calls for 64 agents, diverse proof routes and adversarial checks — danshipper · 2026-07-21
- Users can run FABLE 5, KIMI K3, and Grok 4.5 inside Codex via OpenCodex — iamfakhrealam · 2026-07-21
- Hermes Agent adds built-in Word, Excel, PDF and PowerPoint support — Teknium · 2026-07-21
- Super Proxy open-sources a self-hosted multi-provider LLM gateway with fallback and cost caps — Delicious-Flan88 · 2026-07-21
- Open-source MCP server connects Screener.in to live financial data for LLM research workflows — ashutosh_811 · 2026-07-21
- Marker 2 claims better quality than MinerU and docling while hitting 27 pages/sec — VikParuchuri · 2026-07-21