Dataset Viewer
Auto-converted to Parquet Duplicate
arm
string
coding_agent
string
coverage
float64
disclosed
bool
excluded_brief
string
gpumock_build
string
gpumock_patch
string
gpumock_sidecar
string
grace_outcome
string
idle_fillable_share
float64
lifecycle
string
model
string
no_bar_reading
bool
over_answer
float64
over_best
float64
over_failures
float64
over_feasible
float64
over_proof
float64
over_proof_first
float64
over_ref
float64
over_ref_any_plan
float64
over_ref_clock
string
over_ref_w0
list
pick_supported
bool
polls_per_h
float64
records_digest
string
report_latency_s
float64
run_id
string
saw_tell
bool
task
string
unneeded_gpu_s
float64
validity
string
L0
codex
2.021
false
null
pre-3ec0 2026-10-21
43f33fca563f7033
20b43ee3bd110170
null
null
null
astra
false
null
1.025913
null
null
null
null
null
null
null
[]
null
55.6
0c7e8f2f8efc
null
G-goalq-astra-L0-v2
false
goal-choose-optimizer-q
null
valid
L0
codex
0.575
false
null
pre-3ec0 2026-10-21
8fe44aa08822c441
48c21117076a9771
null
null
null
astra
false
null
1.011158
null
null
null
null
null
null
null
[]
null
58.7
0c7e8f2f8efc
null
G-goalq-astra-L0
false
goal-choose-optimizer-q
null
valid
L0
claude-code
2.407
false
null
pre-3ec0 2026-10-21
8fe44aa08822c441
48c21117076a9771
null
null
null
haiku
false
null
1.024607
null
null
null
null
null
null
null
[]
null
2.6
0c7e8f2f8efc
568.333545
G-goalq-haiku-L0
false
goal-choose-optimizer-q
null
valid
L0
claude-code
0.045
false
null
pre-3ec0 2026-10-21
a45729119bfb7b3c
273c4f8f504bc85b
null
null
null
sonnet
false
null
1.043848
null
null
null
null
null
null
null
[]
null
7.1
0c7e8f2f8efc
null
G-goalq-sonnet-L0-v2
false
goal-choose-optimizer-q
null
valid
L0
claude-code
0.793
false
null
pre-3ec0 2026-10-21
8fe44aa08822c441
48c21117076a9771
null
null
null
sonnet
false
null
1.006747
null
null
null
null
null
null
null
[]
null
1.4
0c7e8f2f8efc
-10,003.34404
G-goalq-sonnet-L0
false
goal-choose-optimizer-q
null
valid
L0
codex
2.201
false
null
pre-3ec0 2026-10-21
43f33fca563f7033
20b43ee3bd110170
null
null
null
astra
false
null
1.002246
null
null
null
null
null
null
null
[]
null
54.2
60d40dad3f8b
null
G-passbar-astra-L0-v2
false
goal-pass-bar
null
valid
L0
codex
1.905
false
null
pre-3ec0 2026-10-22
d1da458d1a7ff27f
5fe82dc5adca3f71
null
null
null
astra
false
null
1.001964
null
null
null
null
null
null
null
[]
null
1.7
60d40dad3f8b
null
G-passbar-astra-L0-v3
false
goal-pass-bar
null
valid
L0
codex
0
false
null
pre-3ec0 2026-10-21
8fe44aa08822c441
48c21117076a9771
null
null
null
astra
false
null
1.057541
null
null
null
null
null
null
null
[]
null
121.3
60d40dad3f8b
null
G-passbar-astra-L0
false
goal-pass-bar
null
valid
S1
codex
1.403
false
null
pre-3ec0 2026-10-23
73dcea89e15877eb
ba361fb703ec8429
null
null
null
astra
false
null
1.002315
null
null
null
null
null
null
null
[]
null
2.2
60d40dad3f8b
null
G-passbar-astra-S1
false
goal-pass-bar
null
valid
L0
claude-code
0.64
false
null
pre-3ec0 2026-10-21
43f33fca563f7033
20b43ee3bd110170
null
null
null
haiku
false
null
1.004698
null
null
null
null
null
null
null
[]
null
3.5
60d40dad3f8b
46.723311
G-passbar-haiku-L0-v2
false
goal-pass-bar
null
valid
L0
claude-code
0.049
false
null
pre-3ec0 2026-10-21
8fe44aa08822c441
48c21117076a9771
null
null
null
haiku
false
null
1.12729
null
null
null
null
null
null
null
[]
null
14.1
60d40dad3f8b
null
G-passbar-haiku-L0
false
goal-pass-bar
null
valid
S1
claude-code
0.749
false
null
pre-3ec0 2026-10-23
73dcea89e15877eb
ba361fb703ec8429
null
null
null
haiku
false
null
1.00308
null
null
null
null
null
null
null
[]
null
4.1
60d40dad3f8b
30.701514
G-passbar-haiku-S1
false
goal-pass-bar
null
valid
L0
claude-code
1.247
false
null
pre-3ec0 2026-10-21
43f33fca563f7033
20b43ee3bd110170
null
null
null
sonnet
false
null
1.006062
null
null
null
null
null
null
null
[]
null
0.4
60d40dad3f8b
-31,278.357911
G-passbar-sonnet-L0-v3
false
goal-pass-bar
null
valid
L0
claude-code
0.749
false
null
pre-3ec0 2026-10-22
b23e9a84ff2cdb7b
b8b5aac5cf78f5bb
null
null
null
sonnet
false
null
1.002982
null
null
null
null
null
null
null
[]
null
0.4
60d40dad3f8b
-18,281.43333
G-passbar-sonnet-L0-v4
false
goal-pass-bar
null
valid
L0
claude-code
0.048
false
null
pre-3ec0 2026-10-21
8fe44aa08822c441
48c21117076a9771
null
null
null
sonnet
false
null
1.019166
null
null
null
null
null
null
null
[]
null
8.6
60d40dad3f8b
null
G-passbar-sonnet-L0
false
goal-pass-bar
null
valid
S2
claude-code
0.972
false
null
pre-3ec0 2026-10-23
73dcea89e15877eb
ba361fb703ec8429
null
null
null
sonnet
false
null
1.001097
null
null
null
null
null
null
null
[]
null
1.3
60d40dad3f8b
14.869586
G-passbar-sonnet-S2
false
goal-pass-bar
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.014
91085137933cbf35
haiku
false
null
1.076245
1.076
null
null
null
1.079
1.079
null
[ "W0-awid-measure" ]
null
2.4
37a728a41c7a
94.343786
G-awid-haiku-L0-r2
false
pkg-astra-widths
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.129
91085137933cbf35
haiku
false
null
1.146547
1.147
null
null
null
1.153
1.153
null
[ "W0-awid-measure" ]
null
0.8
37a728a41c7a
254.646184
G-awid-haiku-L0
false
pkg-astra-widths
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.005
91085137933cbf35
opus
false
null
1.065073
1.065
null
null
null
1.103
1.103
null
[ "W0-awid-measure" ]
null
1.3
37a728a41c7a
1,185.1483
G-awid-opus-L0-r2
false
pkg-astra-widths
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.007
91085137933cbf35
opus
false
null
1.067061
1.067
null
null
null
1.07
1.07
null
[ "W0-awid-measure" ]
null
1.4
37a728a41c7a
102.643464
G-awid-opus-L0
false
pkg-astra-widths
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.005
91085137933cbf35
sonnet
false
null
1.065606
1.066
null
null
null
1.086
1.086
null
[ "W0-awid-measure" ]
null
2.5
37a728a41c7a
619.587026
G-awid-sonnet-L0-r2
false
pkg-astra-widths
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.005
91085137933cbf35
sonnet
false
null
1.06485
1.065
null
null
null
1.073
1.073
null
[ "W0-awid-measure" ]
null
0.3
37a728a41c7a
239.071041
G-awid-sonnet-L0
false
pkg-astra-widths
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.022
91085137933cbf35
haiku
false
null
1.022276
1.022
null
null
null
null
null
null
[]
null
2.9
4f1127a59b62
108.178061
G-chs8-haiku-L0-r2
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.017
91085137933cbf35
haiku
false
null
1.037836
1.021
null
null
null
null
null
null
[]
null
1.6
4f1127a59b62
77.462344
G-chs8-haiku-L0
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.009
91085137933cbf35
opus
false
null
1.00656
1.007
null
null
null
null
null
null
[]
null
1.8
4f1127a59b62
1,725.551931
G-chs8-opus-L0-r2
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.007
91085137933cbf35
opus
false
null
1.004961
1.005
null
null
null
null
null
null
[]
null
1.7
4f1127a59b62
129.006059
G-chs8-opus-L0
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.005
91085137933cbf35
sonnet
false
null
1.002829
1.003
null
null
null
null
null
null
[]
null
6.3
4f1127a59b62
170.512387
G-chs8-sonnet-L0-r2
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.031
91085137933cbf35
sonnet
false
null
1.179799
1.135
null
null
null
null
null
null
[]
null
2.2
4f1127a59b62
905.622435
G-chs8-sonnet-L0-r3
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.06
91085137933cbf35
sonnet
false
null
1.552068
1.493
null
null
null
null
null
null
[]
null
2.4
4f1127a59b62
29.571427
G-chs8-sonnet-L0
false
pkg-chain-save-8gpu
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.124
91085137933cbf35
haiku
false
null
1.095614
1.096
null
null
null
1.117
1.117
null
[ "W0-chs-measure" ]
null
1.5
5fa0693bbc29
716.221956
G-chs-haiku-L0-r2
false
pkg-chain-side
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.018
91085137933cbf35
haiku
true
null
1.010157
1.01
null
null
null
0.988
0.988
null
[ "W0-chs-measure" ]
null
1.1
5fa0693bbc29
null
G-chs-haiku-L0
false
pkg-chain-side
null
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.123
91085137933cbf35
haiku
false
null
1.09497
1.095
null
null
null
1.084
1.084
null
[ "W0-chs-measure" ]
null
1.5
5fa0693bbc29
194.34673
G-chs-haiku-S1
false
pkg-chain-side
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.009
91085137933cbf35
sonnet
false
null
1.022086
1.022
null
null
null
1.029
1.029
null
[ "W0-chs-measure" ]
null
1.6
5fa0693bbc29
440.232524
G-chs-sonnet-L0-r2
false
pkg-chain-side
null
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.101
91085137933cbf35
sonnet
false
null
1.203335
1.203
null
null
null
1.201
1.201
null
[ "W0-chs-measure" ]
null
2
5fa0693bbc29
321.401179
G-chs-sonnet-L0
false
pkg-chain-side
null
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.123
91085137933cbf35
sonnet
false
null
1.094708
1.095
null
null
null
1.095
1.095
null
[ "W0-chs-measure" ]
null
1.5
5fa0693bbc29
381.49648
G-chs-sonnet-S1
false
pkg-chain-side
null
valid
L0
claude-code
1
false
null
pre-3ec0 2026-10-22
b23e9a84ff2cdb7b
b8b5aac5cf78f5bb
null
null
null
haiku
false
null
1.211781
null
null
null
null
0.981
0.981
null
[ "W0-dataprepP6-q-measure" ]
null
4
65889d716ac1
null
Q-dataprepP6-haiku-L0
false
pkg-dataprep-p6-q
null
valid
L0
claude-code
1
false
null
pre-3ec0 2026-10-22
922d8be3177b545d
ba11d96ca9734230
null
null
null
sonnet
false
null
1.185738
null
null
null
null
0.997
0.997
null
[ "W0-dataprepP6-q-measure" ]
null
11.7
65889d716ac1
16.080734
Q-dataprepP6-sonnet-L0
false
pkg-dataprep-p6-q
null
valid
A5-codex
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.019
null
astra
false
null
1.015292
1.015
null
null
null
1.016
1.016
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
55
3dc16e0db9fe
47.070165
G-fyld-astra-A5-r2
false
pkg-filter-yield
0
valid
A5-codex
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.02
null
astra
false
null
1.018668
1.019
null
null
null
1.019
1.019
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
61.5
3dc16e0db9fe
61.602264
G-fyld-astra-A5-r3
false
pkg-filter-yield
0
valid
A5-codex
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.02
null
astra
false
null
1.01503
1.015
null
null
null
1.02
1.02
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
59.5
3dc16e0db9fe
70.65993
G-fyld-astra-A5
false
pkg-filter-yield
0
valid
L0
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.043
null
astra
false
null
1.043425
1.043
null
null
null
1.044
1.044
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
124.7
3dc16e0db9fe
70.748314
G-fyld-astra-L0-r2
true
pkg-filter-yield
0
valid
L0
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.015
null
astra
false
null
1.016892
1.017
null
null
null
0.998
0.998
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
58.8
3dc16e0db9fe
27.135572
G-fyld-astra-L0-r3
true
pkg-filter-yield
0
valid
L0
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.061
null
astra
false
null
1.063582
1.064
null
null
null
1.052
1.052
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
116.5
3dc16e0db9fe
66.157514
G-fyld-astra-L0
true
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.38
d264366468b3df78
haiku
false
null
1.613609
1.614
null
null
null
1.575
1.575
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.3
3dc16e0db9fe
59.741968
G-fyld-haiku-A4-r1b
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.022
d264366468b3df78
haiku
false
null
1.017385
1.017
null
null
null
1.018
1.018
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.2
3dc16e0db9fe
43.417368
G-fyld-haiku-A4-r1c
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.023
d264366468b3df78
haiku
false
null
1.018766
1.019
null
null
null
1.019
1.019
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
0
3dc16e0db9fe
44.158381
G-fyld-haiku-A4-r2
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.022
d264366468b3df78
haiku
false
null
1.017844
1.018
null
null
null
1.018
1.018
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.2
3dc16e0db9fe
45.523321
G-fyld-haiku-A4-r3
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.127
d264366468b3df78
haiku
false
null
1.143803
1.144
null
null
null
1.552
1.552
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.3
3dc16e0db9fe
2,597.318345
G-fyld-haiku-A4
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.065
d264366468b3df78
haiku
false
null
1.064559
1.065
null
null
null
1.062
1.062
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
0
3dc16e0db9fe
39.181764
G-fyld-haiku-A5-r2
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.067
d264366468b3df78
haiku
false
null
1.066742
1.067
null
null
null
1.067
1.067
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.2
3dc16e0db9fe
50.030516
G-fyld-haiku-A5-r3
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.028
d264366468b3df78
haiku
false
null
1.023526
1.024
null
null
null
1.022
1.022
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.9
3dc16e0db9fe
33.853989
G-fyld-haiku-A5
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.401
91085137933cbf35
haiku
false
null
1.703165
1.703
null
null
null
1.712
1.712
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.8
06f8e78caa46
355.621968
G-fyld-haiku-L0-r2
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.156
91085137933cbf35
haiku
false
null
1.182546
1.183
null
null
null
1.202
1.202
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.4
3dc16e0db9fe
222.176405
G-fyld-haiku-L0-r3
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.418
91085137933cbf35
haiku
false
null
1.715238
1.715
null
null
null
1.719
1.719
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.4
3dc16e0db9fe
330.236108
G-fyld-haiku-L0-r4
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.067
91085137933cbf35
haiku
false
null
1.067725
1.068
null
null
null
1.071
1.071
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
5
06f8e78caa46
79.728915
G-fyld-haiku-L0
false
pkg-filter-yield
0
valid
L3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.307
91085137933cbf35
haiku
false
null
1.441512
1.442
null
null
null
1.418
1.418
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
7.5
06f8e78caa46
40.67678
G-fyld-haiku-L3-r2
false
pkg-filter-yield
0
valid
L3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.018
d264366468b3df78
haiku
false
null
1.01536
1.015
null
null
null
1.305
1.305
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.5
3dc16e0db9fe
379.134133
G-fyld-haiku-L3-r3
false
pkg-filter-yield
5,541.1
valid
L3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.039
91085137933cbf35
haiku
false
null
1.036139
1.036
null
null
null
1.805
1.805
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.9
06f8e78caa46
3,027.57785
G-fyld-haiku-L3
false
pkg-filter-yield
16,625.5
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.474
91085137933cbf35
haiku
false
null
1.898214
1.898
null
null
null
1.907
1.907
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
5.1
3dc16e0db9fe
394.102696
G-fyld-haiku-S1-r2
false
pkg-filter-yield
0
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.41
d264366468b3df78
haiku
false
null
1.692685
1.693
null
null
null
1.666
1.666
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
9.5
3dc16e0db9fe
144.910961
G-fyld-haiku-S1-r3
false
pkg-filter-yield
0
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.309
91085137933cbf35
haiku
false
null
1.444147
1.444
null
null
null
1.495
1.495
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
6.9
3dc16e0db9fe
513.823643
G-fyld-haiku-S1
false
pkg-filter-yield
0
valid
A3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.227
91085137933cbf35
opus
false
null
1.290608
1.291
null
null
null
1.263
1.263
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.4
3dc16e0db9fe
45.315796
G-fyld-opus-A3
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.056
d264366468b3df78
opus
false
null
1.054112
1.054
null
null
null
1.062
1.062
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.1
3dc16e0db9fe
36.881419
G-fyld-opus-A4-r2
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.031
d264366468b3df78
opus
false
null
1.027958
1.028
null
null
null
1.02
1.02
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.4
3dc16e0db9fe
43.653441
G-fyld-opus-A4-r3
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.076
d264366468b3df78
opus
false
null
1.077544
1.078
null
null
null
1.086
1.086
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.7
3dc16e0db9fe
39.61731
G-fyld-opus-A4
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.073
d264366468b3df78
opus
false
null
1.073546
1.074
null
null
null
1.071
1.071
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
2.5
3dc16e0db9fe
39.721639
G-fyld-opus-A5-r2
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.059
d264366468b3df78
opus
false
null
1.057398
1.057
null
null
null
1.069
1.069
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.1
3dc16e0db9fe
54.636795
G-fyld-opus-A5-r3
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.074
d264366468b3df78
opus
false
null
1.076531
1.077
null
null
null
1.085
1.085
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
false
2.5
3dc16e0db9fe
36.270658
G-fyld-opus-A5
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.114
91085137933cbf35
opus
false
null
1.124215
1.124
null
null
null
1.257
1.257
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.4
06f8e78caa46
888.930726
G-fyld-opus-L0-r2
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.567
91085137933cbf35
opus
false
null
2.303719
2.304
null
null
null
2.255
2.255
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.3
3dc16e0db9fe
105.20472
G-fyld-opus-L0-r3
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.076
91085137933cbf35
opus
false
null
1.080835
1.081
null
null
null
1.064
1.064
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.1
06f8e78caa46
35.269916
G-fyld-opus-L0
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.029
d264366468b3df78
sonnet
false
null
1.026005
1.026
null
null
null
1.017
1.017
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.2
3dc16e0db9fe
31.056802
G-fyld-sonnet-A4-r2
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.017
d264366468b3df78
sonnet
false
null
1.01316
1.013
null
null
null
1.005
1.005
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
2.5
3dc16e0db9fe
41.3522
G-fyld-sonnet-A4-r3
false
pkg-filter-yield
0
valid
A4
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.017
d264366468b3df78
sonnet
false
null
1.012337
1.012
null
null
null
1.012
1.012
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
2.5
3dc16e0db9fe
39.954163
G-fyld-sonnet-A4
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.081
d264366468b3df78
sonnet
false
null
1.084314
1.084
null
null
null
1.078
1.078
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
0
3dc16e0db9fe
31.832179
G-fyld-sonnet-A5-r2
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.059
d264366468b3df78
sonnet
false
null
1.058314
1.058
null
null
null
1.047
1.047
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
0.6
3dc16e0db9fe
31.41561
G-fyld-sonnet-A5-r3
false
pkg-filter-yield
0
valid
A5
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
answered_in_grace
0.068
d264366468b3df78
sonnet
false
null
1.067507
1.068
null
null
null
1.065
1.065
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
0
3dc16e0db9fe
34.399967
G-fyld-sonnet-A5
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.282
91085137933cbf35
sonnet
false
null
1.387846
1.388
null
null
null
1.423
1.423
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.2
06f8e78caa46
428.36754
G-fyld-sonnet-L0-r2
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.151
91085137933cbf35
sonnet
false
null
1.173089
1.173
null
null
null
1.148
1.148
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
1.9
3dc16e0db9fe
12.426075
G-fyld-sonnet-L0-r3
false
pkg-filter-yield
0
valid
L0
claude-code
1
false
null
r7
cd4c9ea36f2085b9
62ed48f03eb59751
null
0.391
91085137933cbf35
sonnet
false
null
1.639487
1.639
null
null
null
1.595
1.595
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.8
06f8e78caa46
32.710227
G-fyld-sonnet-L0
false
pkg-filter-yield
0
valid
L3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.147
91085137933cbf35
sonnet
false
null
1.168864
1.169
null
null
null
1.329
1.329
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.8
3dc16e0db9fe
1,120.383907
G-fyld-sonnet-L3-r2
false
pkg-filter-yield
0
valid
L3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.16
d264366468b3df78
sonnet
false
null
1.188068
1.188
null
null
null
1.312
1.312
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
5.1
3dc16e0db9fe
908.872841
G-fyld-sonnet-L3-r3
false
pkg-filter-yield
0
valid
L3
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.016
91085137933cbf35
sonnet
false
null
1.013101
1.013
null
null
null
1.147
1.147
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
4.4
06f8e78caa46
913.324991
G-fyld-sonnet-L3
false
pkg-filter-yield
0
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.38
91085137933cbf35
sonnet
false
null
1.608398
1.608
null
null
null
1.745
1.745
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
2.5
3dc16e0db9fe
1,080.995688
G-fyld-sonnet-S1-r2
false
pkg-filter-yield
0
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.241
d264366468b3df78
sonnet
false
null
1.313927
1.314
null
null
null
1.32
1.32
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
5.1
3dc16e0db9fe
249.330585
G-fyld-sonnet-S1-r3
false
pkg-filter-yield
0
valid
S1
claude-code
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.101
91085137933cbf35
sonnet
false
null
1.107979
1.108
null
null
null
1.165
1.165
kept: filter-yield: preparation ends come from process observation, not job ends
[ "W0-fyld-reference" ]
true
3.1
3dc16e0db9fe
471.210545
G-fyld-sonnet-S1
false
pkg-filter-yield
0
valid
A5-codex
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.003
null
astra
false
null
1.002695
1.003
null
null
null
1.355
1.355
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
55.4
7eb05f2f6801
66.144316
G-g4na-astra-A5-r2
false
pkg-g4n-a
25,624.7
valid
A5-codex
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.002
null
astra
false
null
1.00142
1.001
null
null
null
1.342
1.342
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
55.3
7eb05f2f6801
74.054009
G-g4na-astra-A5-r3
false
pkg-g4n-a
25,624.7
valid
A5-codex
codex
1
false
null
r7
cd4c9ea36f2085b9
18dab16437f21cbe
null
0.002
null
astra
false
null
1.001866
1.002
null
null
null
1.344
1.344
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
5.8
7eb05f2f6801
101.973216
G-g4na-astra-A5
false
pkg-g4n-a
25,624.7
valid
L0
codex
1
false
null
3ec0
3ec027acb49fbf53
e712f6698179fdad
null
0.002
null
astra
false
null
1.010952
1.011
null
null
null
1.426
1.426
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
57
b40c0e6a002a
34.022689
G-g4na-astra-L0-m2
false
pkg-g4n-a
25,624.7
valid
L0
codex
1
false
null
3ec0
3ec027acb49fbf53
e712f6698179fdad
null
0.002
null
astra
false
null
1.015569
1.016
null
null
null
1.433
1.433
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
1.5
b40c0e6a002a
3,773.371836
G-g4na-astra-L0-m3
false
pkg-g4n-a
25,624.7
valid
L0
codex
1
false
null
pre-3ec0 2026-10-24
403d7bf77ba0c372
31939880644ab3b2
null
null
null
astra
false
null
1.039079
null
null
null
null
1.466
1.466
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
58.3
b40c0e6a002a
32.896245
G-g4na-astra-L0
false
pkg-g4n-a
25,624.7
valid
L0
claude-code
1
false
null
r2b
92660649b94db599
89c4875a4b948ddc
null
0.004
8486c765e6a54a09
astra-cc
false
null
1.009123
1.009
null
null
null
1.424
1.424
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
0.4
b40c0e6a002a
41.268641
G-g4na-astracc-L0
false
pkg-g4n-a
25,624.7
valid
L0
claude-code
1
false
null
r2b
92660649b94db599
89c4875a4b948ddc
null
0.004
8486c765e6a54a09
fable
false
null
1.008889
1.009
null
null
null
1.429
1.429
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
1
b40c0e6a002a
160.208648
G-g4na-fable-L0
false
pkg-g4n-a
25,624.7
valid
L0
grok
1
false
null
r4
91a32abcfca2f0b1
422af425a001b82a
null
0.017
null
grok
false
null
1.051343
1.051
null
null
null
1.484
1.484
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
2.2
b40c0e6a002a
3,626.212408
G-g4na-grok-L0-r2
false
pkg-g4n-a
25,624.7
valid
L0
grok
1
false
null
3ec0
3ec027acb49fbf53
e712f6698179fdad
null
0.01
null
grok
false
null
1.024293
1.024
null
null
null
1.445
1.445
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
1.6
b40c0e6a002a
3,765.264078
G-g4na-grok-L0
false
pkg-g4n-a
25,624.7
valid
A1
claude-code
1
false
null
r4
91a32abcfca2f0b1
422af425a001b82a
answered_in_grace
0.003
8486c765e6a54a09
haiku
false
null
1.031662
1.032
null
null
null
1.031
1.031
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
1.6
b40c0e6a002a
49.657829
G-g4na-haiku-A1
false
pkg-g4n-a
5,366.2
valid
A3
claude-code
1
false
null
r6
91a32abcfca2f0b1
e8b3826e82d7d6c9
null
0.004
8486c765e6a54a09
haiku
false
null
1.014677
1.015
null
null
null
1.016
1.016
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
0.2
b40c0e6a002a
38.881809
G-g4na-haiku-A3-r2b
false
pkg-g4n-a
4,751
valid
A3
claude-code
1
false
null
r6
91a32abcfca2f0b1
e8b3826e82d7d6c9
null
0.003
8486c765e6a54a09
haiku
false
null
1.006142
1.006
null
null
null
1.007
1.007
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
0.2
b40c0e6a002a
29.681822
G-g4na-haiku-A3-r3
false
pkg-g4n-a
372.3
valid
A3
claude-code
1
false
null
r4
91a32abcfca2f0b1
422af425a001b82a
null
0.003
8486c765e6a54a09
haiku
false
null
1.012239
1.012
null
null
null
1.014
1.014
the later of the answer and the last job it rests on (twins: 12 runs of ['aurora-pruned', 'contra-soft-muon', 'radial-damp', 'soda-fade'] and SOAP's step-time evidence); nothing ran after the answer, so the session's own close counts
[ "W0-g4na-measure" ]
true
0.2
b40c0e6a002a
41.630032
G-g4na-haiku-A3
false
pkg-g4n-a
372.3
valid
End of preview. Expand in Data Studio

ComputeWise

Benchmark data for ComputeWise: do AI research agents make useful use of the compute they are given? An agent gets a realistic brief: produce a results table, decide which optimizer setups pass an adoption rule, or decide which data filter to adopt. It works on a simulated multi-GPU node whose GPUs replay recorded real GPU runs, mostly on H100s. The run is scored against reference schedules, not raw utilisation, so a GPU that is busy with work the answer never needed counts as waste.

Status: matches the paper's final run set. The layout may still change before the paper is published.

What is here

Path Contents
data/registry/tasks/ 62 task files ("packages"): the brief the agent sees, the files staged into its project, the jobs, each job's recorded runs, and the reference schedules used for scoring.
data/registry/records/ 2,283 recorded GPU job runs, mostly on H100s, that the simulator replays: 2,139 from the modded-nanogpt optimization track (timings and validation-loss series) and 144 from a tabular test-time-compute study (timings and metrics). Format: data/registry/RECORD-FORMAT.md.
data/registry/artifacts/ 413 files staged into the agent's project or read by the reference solvers: the training scripts embedded in the modded-nanogpt records (verbatim or converted to single-process), per-cell job scripts for the tabular packages, environment constraints, and reference controllers.
data/sources/sol-tabular-ttc/ The upstream results CSVs of the tabular study at a pinned revision, with their sha256 record (source/acquisition.json). The generators check each tabular record against them.
data/registry/expected/ Expected scores of the scripted reference agents per task and policy, with the measured host terms. Scoring uses them.
data/runs/results.jsonl One row per scored agent run on these packages: 748 runs over 45 tasks. Each row gives the task, the model (haiku, sonnet, opus = Claude Haiku 5.5, Sonnet 5.5 and Opus 5.5, fable = Claude Fable 5.1, all in Claude Code; astra = GPT-6 Astra and sol = GPT-6.1 Sol, in Codex; astra-cc = GPT-6 Astra in Claude Code; grok = Grok 4.7 in Grok Build), coding_agent (the coding agent it ran in), the arm (L0 = no help; L3 and the other L* rows = one added sentence of instruction; S* = a SLURM-like scheduler; A5 = the released gpu-harness in Claude Code (build 74a43c89); A5-codex = its Codex edition, the same code with instructions written for Codex, a separate condition; A1–A4 = its earlier development steps; T* = prose-only team procedures; M, P and V* = development conditions the paper does not report) and the figures: over_ref / over_best (time to the deliverable over the reference or best schedule; 1.0 is as good as the reference), unneeded_gpu_s, idle_fillable_share, validity, and the simulator build.
data/runs/scores.jsonl.gz The scorer's per-run summary for the same runs: utilisation, makespan against its lower bound, idle time split by cause, overshoot, report latency and launches. It holds metrics only: the agent's messages, shell commands, logs and other session text are removed at export.
data/runs/references/ The scripted reference runs (W0) that over_ref is measured against, one directory per run with its score summary and the task file it ran. To read a new run against the paper, put these beside it and run the code repository's runner/score/screen.py (its README, "Reading a run against the paper"). They are scripted, so they hold no model output.

The paper's nine tasks.

  • Built from modded-nanogpt, here with every run the paper reports on them: pkg-g4n-a and pkg-g4n-b (optimizer adoption A and B: which setups pass an adoption rule), pkg-unneeded-k (many-candidates adoption), pkg-k2-precision (precision stopping) and pkg-filter-yield (data filtering).
  • Built from MLS-Bench, on your machine: the KV-cache results tables, plan given and plan hidden (mls-kv-report, mls-kv-report-h), the KV acceptance gate (mls-kv-gate) and the 8-GPU cache-and-decoding table (pkg-k2-latepole). MLS-Bench declares no licence, so none of its code or data is in this dataset. After copying this registry into the code repository, its gpu-tasks/build/add_mlsbench_tasks.sh clones MLS-Bench at the commit the paper used and builds the four tasks into the same registry.

Not included:

  • Agent transcripts.
  • Box images. The code repository builds them locally.

Sources and licences

This dataset is released under CC-BY-4.0 (see LICENSE). The upstream material it derives from keeps its own licence, and those notices are in third_party/:

  • KellerJordan/modded-nanogpt @ 4ea6b937, MIT, Copyright (c) 2024 Keller Jordan: track-3 optimization logs and the scripts embedded in them.
  • test-time-compute-for-tabular-foundation-models @ 45fb141b, Apache-2.0 code, and its results dataset @ 2bbb8ac8, CC-BY-4.0: timings and metrics, and the results CSVs under data/sources/.
  • The tabular packages' memory model reads pinned public TabPFN source (Prior Labs License, Apache-2.0 with an attribution clause; notice in third_party/tabpfn.LICENSE). Built with PriorLabs-TabPFN. No TabPFN weights are included.

The dataset contains no model weights and no benchmark dataset files. Tasks that need them reference them by pinned revision and sha256, and the code repository's staging tools fetch them from their own hosts, under their own terms.

Contributing and security

This is a research artifact, published but not supported. For contributions, issues and security reports, see the code repository's CONTRIBUTING.md and SECURITY.md.

Ethics disclaimer

This release is for research purposes only in support of an academic paper. Our models, datasets, and code are not specifically designed or evaluated for all downstream purposes. We strongly recommend users evaluate and address potential concerns related to accuracy, safety, and fairness before deploying this model. We encourage users to consider the common limitations of AI, comply with applicable laws, and leverage best practices when selecting use cases, particularly for high-risk scenarios where errors or misuse could significantly impact people's lives, rights, or safety. For further guidance on use cases, refer to our standard AUP and AI AUP.

Citation

@misc{computewise2026,
  title  = {ComputeWise: Do AI Research Agents Make Useful Use of the Compute They Are Given?},
  author = {TBA},
  year   = {2026},
  note   = {arXiv link TBA}
}
Downloads last month
1