Classement OpenRouter des outils CLI juin 2026 : Top 10 assistants de codage par usage réel et guide de location Mac

Pourquoi les classements Apps CLI OpenRouter battent les GitHub Stars : cinq pièges

OpenRouter tracks not only models but client applications—the CLIs and IDE extensions that route inference. Its Apps leaderboard (openrouter.ai/rankings) reports 7-day rolling token throughput per app, counting every API call the tool makes on behalf of users. That is production traffic, not install counts or star velocity.

Coding now exceeds half of all OpenRouter traffic (OpenRouter + a16z, 100T-token sample). CLI assistants compete on tool-call reliability, model routing flexibility, and daemon uptime—not demo videos. The June 2–8 week shows Hermes Agent alone near 5T tokens, more than many entire model vendors process in the same window.

1. 01

   GitHub stars lag production: A CLI can spike on launch yet fade when tool-call failure rates climb in long Agent sessions. Token volume updates weekly.

2. 02

   IDE extensions hide behind hosts: Cursor and VS Code plugins often share one OpenRouter app ID. Dedicated CLIs like Kilo Code and Claude Code show up as first-class apps with measurable share.

3. 03

   Multi-model routing inflates leaders: Tools that proxy ten models (Hermes, Kilo Code) naturally pull more tokens than single-vendor CLIs—even when daily active users are similar.

4. 04

   Free tiers distort trials: Experiment traffic from zero-cost models can spike a week; pair rankings with your own cost dashboard before committing team-wide.

5. 05

   Host OS matters for half the Top 10: Claude Code, Goose, and several Agent frameworks expect macOS Keychain, launchd, or Xcode paths—Linux-only hosts add integration tax regardless of model choice.

Top 10 assistants de codage CLI juin 2026 sur OpenRouter

The full Apps category for terminal and IDE coding assistants during June 2–8, 2026 includes the ten tools below. Three publish weekly token totals on the public leaderboard; the rest appear in the category roster with qualitative positioning based on OpenRouter metadata and upstream docs.

Hermes Agent's 4.94T volume reflects daemon-style workloads: Gateway processes, Telegram bots, and scheduled Skills that fire across every hour of the week. Claude Code's lower token count still maps to high ARPU teams—Anthropic billing often runs direct, not only through OpenRouter. Kilo Code sits in the middle: broad model choice with IDE-native UX drives steady multi-provider traffic.

Runbook en six étapes : choisir un CLI, router les modèles, déployer sur Mac

App rankings shift weekly. This runbook turns June 2026 leaderboard data into a reproducible team workflow—from tool selection through OpenRouter keys to a production Mac host.

1. 01

   Pull the Apps tab every Monday: Open openrouter.ai/rankings, switch to Apps, and log CLI Top 10 moves plus token totals for Hermes, Kilo Code, and Claude Code.

2. 02

   Match CLI to workflow shape: Daemon Agents → Hermes; IDE pair programming → Kilo Code or Cline; Anthropic-only enterprise → Claude Code; git-centric diffs → Aider.

3. 03

   Provision OpenRouter keys with app tags: Create per-environment keys labeled by CLI app so billing dashboards attribute spend correctly when you A/B tools.

4. 04

   Regression on a fixed repo subset: Weekly, run the same ten issues through your shortlist. Track tool-call failures and diff acceptance rate—not just latency.

5. 05

   Deploy on a matrix-matched Mac: Use Section 04 RAM tiers. Install CLI via upstream docs; store API keys in Keychain; register launchd plist for daemons.

6. 06

   Validate 7-day uptime before team rollout: Let Hermes or Claude Code run one full week on the rental Mac. Compare token logs against OpenRouter dashboard; only then migrate the team off personal laptops.

{
  "weekly_review": "2026-06-08",
  "cli_stack": {
    "hermes_gateway": {
      "openrouter_api_key": "sk-or-v1-...",
      "default_model": "openrouter/deepseek/deepseek-v4-flash",
      "host": "mac-mini-m4-24gb"
    },
    "claude_code": {
      "anthropic_api_key": "sk-ant-...",
      "openrouter_fallback": "openrouter/anthropic/claude-sonnet-4.6"
    },
    "kilo_code": {
      "openrouter_api_key": "sk-or-v1-...",
      "primary_model": "openrouter/deepseek/deepseek-v4-flash"
    }
  },
  "monthly_cap_usd": 1200
}

Trois questions que les lecteurs posent le plus