Act on this, mostly signal

Summonr 0.1.0 open-source coding agent claims 85.8% on Terminal-Bench

What to do

Run Summonr in a container, since it ships without a sandbox, and check its benchmark claim yourself.

Why

The curl install claims 85.8% resolution versus Codex at 79.0%, and token savings are self-reported.

What the research found

Summonr 0.1.0 is an open-source coding agent with a curl install script claiming 85.8% resolution on Terminal-Bench 2.1 versus Codex 79.0%, with self-reported token and memory savings, and no sandbox so engineers should run it in a container.

Claims and how well the evidence supports them

Summonr 0.1.0 resolved 85.8% on Terminal-Bench 2.1 vs Codex 79.0% 99%
Summonr uses 3x fewer input tokens per solved task than Codex 98%
Summonr has no sandbox and no permission prompts for reversible actions 83%
Summonr installs via curl script and is open source 95%

Evidence

summonr.dev https://summonr.dev/

Actionability rated medium.