Latevsminion
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Late
mlhher · Terminal agent
#49OSS
Panel
6.42 spec wins
- Reliability
- 5.7
- Usefulness
- 6.3
- Cost
- 8.3
- Longevity
- 5.2
“It is primarily developed inside itself, which is either dogfooding or a hostage situation.”
minion
Sentdex · Terminal agent
#50OSS
Panel
6.50 spec wins
- Reliability
- 5.8
- Usefulness
- 6.5
- Cost
- 8.7
- Longevity
- 4.8
“It renders the model's separate reasoning stream as a thinking block, so you can watch it talk itself into the bug.”
Spec by spec
| Spec | Late | minion |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local | local |
| Platforms | macos, linux, windows | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | No | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | No | No |
| Browser control | No | No |
| Sandboxed execution | No | No |
| Multi-agent orchestration | YesSubagents are enforced rather than optional: the orchestrator plans and spawns one ephemeral subagent per atomic edit. | No |
| Headless / CI mode | No | No |
| Models | ||
| Backbone | local models, DeepSeek, any frontier model | any OpenAI-compatible endpoint, llama.cpp, vLLM, SGLang, OpenAI, Z.ai |
| Bring your own model | Yes | Yes |
| Local models | YesThe stated goal is running real work on consumer hardware with tiny local models. | YesMINION_BASE_URL points at any OpenAI-compatible server; local llama.cpp, vLLM and SGLang are the primary target. |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | $0/mo | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | NOASSERTION | MIT |
| GitHub stars | 440 | 337 |
Which one would each critic pick
| Critic | Late | minion | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.8 | 6.8 | no preference |
| El Crítico | 6.0 | 5.8 | Late — It runs shell commands and edits files with no sandbox and no git integration, so the thing making the edit exits before you can ask it anything. |
| El Profesor | 7.0 | 7.3 | minion — The 625-token figure is measured precisely: a bare greeting, a system prompt, five tool schemas and the chat-template framing, which is a definition most claims of this kind lack. |
| La Inversora | 5.8 | 5.8 | no preference |
| La Jefa | 5.3 | 5.3 | no preference |
| El Hacker | 7.5 | 8.0 | minion — MIT, MINION_BASE_URL points at llama.cpp, vLLM or SGLang, and when a server has no native tool calling it falls back to parsing the text, which is the detail that gives it away. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.