Terminal-Bench 2.0 on claude-opus-4-5
89 tasks, 1 attempt, open network. Updated 25 Oct 2026.
Compare
Pick two harnesses to compare them task by task.
mux against OpenCode, task by task
Over the 89 tasks both ran, 1 attempt each
mux minus OpenCode, pass rate
+6.7 points
95% interval −2.2 to +15.7
The interval includes zero, so the paired test can't tell the two apart.
Where the tasks landed
- 12Only mux solved
- 40Both solved
- 6Only OpenCode solved
- 31Neither solved
Of 89 tasks, 12 only mux solved, 6 only OpenCode solved, 40 both solved and 31 neither did.
| Task | Reward | Tokens | Time | |||
|---|---|---|---|---|---|---|
| mux | OpenCode | mux | OpenCode | mux | OpenCode | |
break-filter-js-from-html | 1 | 0 | — | — | 8.2 min | 20.0 min |
build-pmars | 1 | 0 | 4,799 | — | 1.9 min | 2.1 min |
cancel-async-tasks | 1 | 0 | 2,307 | — | 45 s | 22 s |
configure-git-webserver | 1 | 0 | — | — | 2.0 min | 2.6 min |
count-dataset-tokens | 1 | 0 | — | — | 2.0 min | 1.9 min |
db-wal-recovery | 1 | 0 | 3,046 | — | 3.4 min | 6.4 min |
dna-assembly | 0 | 1 | — | — | 15.6 min | 14.7 min |
extract-moves-from-video | 0 | 1 | — | — | 17.1 min | 21.3 min |
financial-document-processor | 1 | 0 | — | — | 2.8 min | 1.2 min |
largest-eigenval | 0 | 1 | — | — | 3.5 min | 6.5 min |
overfull-hbox | 0 | 1 | — | — | 3.7 min | 5.0 min |
path-tracing-reverse | 1 | 0 | — | — | 13.6 min | 19.1 min |
pypi-server | 0 | 1 | 7,463 | — | 3.4 min | 2.7 min |
pytorch-model-recovery | 1 | 0 | — | — | 9.7 min | 3 s |
qemu-startup | 1 | 0 | 3,945 | — | 3.5 min | 15.0 min |
rstan-to-pystan | 0 | 1 | — | — | 30.0 min | 12.1 min |
sanitize-git-repo | 1 | 0 | — | — | 3.2 min | 1.9 min |
sparql-university | 1 | 0 | — | — | 1.6 min | 37 s |