mini-SWE-agentvsQwen Code
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
mini-SWE-agent
SWE-agent (Princeton and Stanford) · Terminal agent
#8OSS
Panel
6.91 spec wins
- Reliability
- 6.3
- Usefulness
- 5.5
- Cost
- 8.8
- Longevity
- 7.0
“One hundred lines of Python, which is fewer than most competitors spend on their pricing page.”
Qwen Code
Alibaba Cloud (Qwen) · Terminal agent
#20OSSMCP
Panel
6.84 spec wins
- Reliability
- 6.5
- Usefulness
- 6.8
- Cost
- 7.3
- Longevity
- 6.3
“Ships a Java SDK, which tells you precisely which enterprises Alibaba expects to pay for it.”
Spec by spec
| Spec | mini-SWE-agent | Qwen Code |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local, sandbox | local |
| Platforms | macos, linux | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | No | Yes |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | Yes |
| Browser control | No | YesThe first-party @qwen-code/chrome-bridge extension tunnels cdp_* frames from the `qwen serve` daemon into chrome.debugger so the agent drives a real Chrome tab, but console/network/screenshot/click tools need an external CDP MCP adapter set via QWEN_CDP_MCP_COMMAND, since none is bundled. |
| Sandboxed execution | Yes | YesTwo real isolation modes - macOS Seatbelt via sandbox-exec (default `permissive-open` profile) and container-based Docker/Podman with a published sandbox image (https://qwenlm.github.io/qwen-code-docs/en/users/features/sandbox/). |
| Multi-agent orchestration | No | Yes |
| Headless / CI mode | Yes | Yes |
| Models | ||
| Backbone | any via litellm | Qwen3-Coder, OpenAI-compatible, Claude, Gemini |
| Bring your own model | Yes | Yes |
| Local models | Yes | Yes |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | n/a | n/a |
| Free tier | Yes | No |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | Apache-2.0 |
| GitHub stars | 8,410 | 28,432 |
Which one would each critic pick
| Critic | mini-SWE-agent | Qwen Code | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.8 | 7.0 | Qwen Code — Pick Qwen Code if you want a Claude Code-shaped terminal with plugins for VS Code, JetBrains and Zed and your choice of model; pick Claude Code if you want the best model with zero setup. |
| El Crítico | 6.5 | 6.3 | mini-SWE-agent — There is no procedure for anything beyond a shell command, opening a pull request is tell the LM to figure it out, and without a sandbox flag every command lands on the host. |
| El Profesor | 8.0 | 6.0 | mini-SWE-agent — The reference harness for the SWE-bench bash-only leaderboard: one bash action per turn via subprocess.run, a linear history, no tool-calling API, and Gemini 3 Pro reported above 74% on Verified. |
| La Inversora | 6.8 | 6.8 | no preference |
| La Jefa | 5.0 | 5.8 | Qwen Code — No seat and nothing to sign; the Coding Plan is a weekly quota on Alibaba Cloud with a Beijing or an international endpoint, and legal will ask which one sixty laptops point at. |
| El Hacker | 8.5 | 8.8 | Qwen Code — Apache-2.0, OPENAI_BASE_URL pointed at my own box, modelProviders in settings.json, and MCP servers with a command or an httpUrl; the whole thing is mine to fork. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.