Agent SvsLegion
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Agent S
Simular · Agent framework
OSS
Panel
6.44 spec wins
- Reliability
- 6.0
- Usefulness
- 6.5
- Cost
- 6.5
- Longevity
- 6.7
“Surpassed the human baseline on OSWorld, then asked you to install tesseract with Homebrew first.”
Legion
Software Mansion · Agent framework
OSS
Panel
6.52 spec wins
- Reliability
- 6.0
- Usefulness
- 6.2
- Cost
- 8.0
- Longevity
- 5.8
“It writes Lua to get work done in Elixir, which is two functional languages agreeing that a third one should take the blame.”
Spec by spec
| Spec | Agent S | Legion |
|---|---|---|
| Architecture | ||
| Category | Agent framework | Agent framework |
| Runs | local | local |
| Platforms | macos, linux, windows | macos, linux |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | No | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | No |
| Multi-file edits | No | Yes |
| Git operations | No | No |
| Browser control | Yes | No |
| Sandboxed execution | No | YesLegion runs generated Lua or Elixir in its own sandbox; the README does not describe a container runtime. |
| Multi-agent orchestration | Yes | No |
| Headless / CI mode | No | No |
| Models | ||
| Backbone | any | multiple providers |
| Bring your own model | Yes | Yes |
| Local models | No | No |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | n/a | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | Apache-2.0 | MIT |
| GitHub stars | 12,494 | 220 |
Which one would each critic pick
| Critic | Agent S | Legion | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.8 | 6.8 | no preference |
| El Crítico | 6.3 | 6.0 | Agent S — The prerequisites say plainly that the agent runs Python code to control your computer, and the board records no container sandbox, so the blast radius is your desktop. |
| El Profesor | 6.8 | 7.8 | Legion — One evaluation can filter, branch and loop, so a sequence that costs a tool-calling agent a model round trip per step collapses into a single call. The claim is structural and checkable. |
| La Inversora | 6.3 | 6.3 | no preference |
| La Jefa | 5.0 | 5.5 | Legion — Nothing to license and nothing to govern: this is a Hex dependency my engineers add to a mix file, so what I am approving is a service my own team will operate. |
| El Hacker | 7.5 | 6.8 | Agent S — Apache-2.0, pip install gui-agents, keys come from environment variables, and vLLM is a supported inference path so the whole stack can run on hardware I own. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.