AutoMode compared with other routers
AutoMode against effortless, jev-router and claude-router for Claude Code: features, how each runs, measured cost on the same sessions, and trade-offs.
In short
Four open source routers work with Claude Code: AutoMode, effortless, jev-router and claude-router. effortless sets the effort for each prompt and keeps the model you picked. jev-router routes the model through a local proxy and also runs on Codex. claude-router lets you choose the classifier, including a local Ollama model. AutoMode sets both the model and the effort, needs no extra key, adds a quality floor for risky work, always reviews risky changes, and runs a plan, build, verify and review workflow for big jobs. It also checks the prompt cache cost before every model switch. On the same five mixed sessions, AutoMode cost $2.01 and effortless $3.87, and both passed 20 of 20 steps. Plain Opus cost $4.14. jev-router and claude-router were audited from source, not run, and neither publishes a measured saving. AutoMode's effort rules are ported from effortless, so the projects are related.
Four routers for Claude Code. Feature rows come from each project's README, read on 2026-10-30. The cost rows come from our own benchmark run on 2026-11-04. These projects change quickly, so check their pages too. If a cell is wrong, open an issue.
AutoMode is not independent of these projects. Its effort rules and bar techniques are ported from effortless, and its classifier contract and switch payback are adapted from claude-router and jev-router.
| AutoMode | effortless | jev-router | claude-router | |
|---|---|---|---|---|
| Effort per prompt | Yes | Yes | Not described | Yes |
| Model per prompt | Yes, on by default | Off by default, marked experimental | Yes | Yes |
| Extra credential | None | None. A Jev key is optional. | Jev key | Depends on the classifier. Ollama needs none. |
| How it runs | Claude Code mod | Claude Code mod | Local proxy around the CLI | Claude Code mod |
| CLIs | Claude Code | Claude Code | Claude Code and Codex | Claude Code |
| Multi-stage workflow | Plan, build, verify, review | Not described | Not described | Not described |
| Mandatory review for risky work | Yes. Opus when the diff shows critical risk | Not described | Not described | Not described |
| Prompt cache | Countdown on the bar, cost check before a switch | Countdown on the bar, compact and handoff offers | Refuses a downgrade that wastes more cache than it saves | Cache-cost rules in its policy |
| Cost result | Measured: 51% less than Opus in mixed sessions, 45% less on 30 coding tasks, same pass rate | Measured by us: 7% less than Opus in the same sessions | None found | "Savings are not measured" |
"Not described" means the README does not mention it. It does not prove the feature is absent.
Measured on the same sessions
| Arm | Passed | Questions | Coding tasks | Total | vs Opus |
|---|---|---|---|---|---|
| AutoMode | 20/20 | $0.08 | $1.93 | $2.01 | −51% |
| Plain Sonnet high | 20/20 | $0.24 | $1.90 | $2.14 | −48% |
| effortless, on Opus | 20/20 | $0.43 | $3.44 | $3.87 | −7% |
| Plain Opus | 20/20 | $0.50 | $3.64 | $4.14 | — |
effortless sets the effort and keeps the model. AutoMode also routes the model: questions go to Haiku and coding to Sonnet. jev-router and claude-router were audited from source, not run.
Other tools with similar names
- ClaudeRouter classifies each prompt and delegates to Haiku, Sonnet or Opus. It installs from npm.
- claude-model-router tracks token use and routes sub-tasks to subagents, rather than the main prompt.
- Claude Code Router does a different job. It connects coding agents to the providers you choose, then routes and fails over between them.
Where the others are ahead
- effortless is simpler. It does one job.
- jev-router also runs on Codex. AutoMode runs only in Claude Code.
- claude-router lets you choose the classifier, including a local Ollama model. AutoMode's judge always runs on Haiku, through your Claude login.
Longer comparisons: AutoMode vs effortless and AutoMode vs Claude Code Router.
Last updated on
Run the benchmark yourself
Run AutoMode's benchmark harness yourself: a dry run, a study on your own plan, grading and reports, plus free tools that read your own transcripts.
Troubleshooting
Fix common AutoMode problems: nothing changes after install, the mod is not listed, the bar is idle or missing, a hook is skipped, and filing a bug.