Terminal-Bench 2.0 on GPT-5.3-Codex

89 tasks, 5 attempts, open network. Updated 25 Oct 2026.

Pick two harnesses to compare them task by task.

mux against droid, task by task

Over the 89 tasks both ran, 5 attempts each

mux minus droid, pass rate

−2.7 points

95% interval −8.3 to +2.7

The interval includes zero, so the paired test can't tell the two apart.

Where the tasks landed

Of 89 tasks, 7 only mux solved, 7 only droid solved, 60 both solved and 15 neither did.

Rewards differ (31) Every task (89)

mux and droid, task by task
Task	Reward		Tokens		Time
Task	mux	droid	mux	droid	mux	droid
`mcmc-sampling-stan`	1	0	1.0M	—	19.1 min	10.0 min
`polyglot-c-py`	0.2	1	444k	—	7.1 min	5.8 min
`polyglot-rust-c`	0	0.8	248k	—	7.0 min	8.5 min
`configure-git-webserver`	0	0.6	8,580	—	1.1 min	3.8 min
`mailman`	0.4	1	—	—	30.0 min	11.9 min
`winning-avg-corewars`	0.4	1	217k	—	20.1 min	19.0 min
`adaptive-rejection-sampler`	0.6	1	399k	—	7.6 min	8.0 min
`cancel-async-tasks`	0.6	1	13k	—	1.2 min	2.0 min
`compile-compcert`	0.4	0.8	1.7M	—	11.5 min	15.9 min
`dna-assembly`	0.4	0	637k	—	27.7 min	8.9 min
`dna-insert`	0.8	0.4	209k	—	18.7 min	3.9 min
`gcode-to-text`	0	0.4	1.7M	—	14.9 min	11.4 min
`make-mips-interpreter`	0.4	0.8	2.6M	—	15.5 min	14.7 min
`model-extraction-relu-logits`	0.8	0.4	282k	—	9.1 min	10.1 min
`mteb-leaderboard`	1	0.6	3.2M	—	15.6 min	15.8 min
`overfull-hbox`	0.8	0.4	242k	—	3.4 min	3.1 min
`protein-assembly`	0.6	1	611k	—	23.5 min	7.0 min
`qemu-alpine-ssh`	0.4	0	457k	—	14.5 min	15.0 min
`torch-tensor-parallelism`	0.8	0.4	199k	—	4.2 min	5.4 min
`extract-elf`	0.6	0.8	344k	—	6.7 min	5.9 min
`tune-mjcf`	0.6	0.8	93k	—	10.3 min	9.3 min
`extract-moves-from-video`	0.2	0.4	118k	—	24.6 min	28.0 min
`mteb-retrieve`	0.2	0	60k	—	1.9 min	2.2 min
`torch-pipeline-parallelism`	0.2	0	217k	—	6.8 min	7.1 min
`train-fasttext`	0	0.2	—	—	30.1 min	54.9 min
`chess-best-move`	0.8	1	860k	—	9.0 min	3.2 min
`db-wal-recovery`	0.6	0.4	1.1M	—	10.0 min	6.9 min
`feal-differential-cryptanalysis`	1	0.8	315k	—	9.7 min	6.0 min
`path-tracing`	0.8	1	2.3M	—	16.0 min	8.4 min
`query-optimize`	0.6	0.4	255k	—	12.8 min	13.0 min
`schemelike-metacircular-eval`	0.8	1	860k	—	10.9 min	14.4 min