Dataset Viewer
Auto-converted to Parquet Duplicate
agent_order
listlengths
5
5
best_agents
listlengths
1
5
dataset
stringclasses
6 values
difficulty
stringclasses
3 values
domain
stringclasses
3 values
is_tie
bool
2 classes
mean_reward
listlengths
5
5
n_correct
listlengths
5
5
n_empty
listlengths
5
5
n_error
listlengths
5
5
pool_version
stringclasses
1 value
repetitions
int64
3
3
split
stringclasses
3 values
task_group
stringclasses
2 values
task_id
stringlengths
16
25
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
opencompass/AIME2025
hard
math
true
[ 1, 1, 0.6666666666666666, 0, 1 ]
[ 3, 3, 2, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
held_out
aime2025_i/test/000002
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a" ]
opencompass/AIME2025
hard
math
false
[ 0.3333333333333333, 0, 0, 0, 0 ]
[ 1, 0, 0, 0, 0 ]
[ 0, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
held_out
aime2025_ii/test/000007
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
allenai/ai2_arc
medium
reasoning
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
arc_challenge/test/000009
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
allenai/ai2_arc
medium
reasoning
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
arc_challenge/test/000011
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
allenai/ai2_arc
medium
reasoning
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
arc_challenge/test/000031
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
allenai/ai2_arc
medium
reasoning
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
arc_challenge/test/000051
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
allenai/ai2_arc
medium
reasoning
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
arc_challenge/test/000065
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
allenai/ai2_arc
medium
reasoning
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
arc_challenge/test/000083
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
openai/gsm8k
easy
math
true
[ 1, 1, 0, 0.6666666666666666, 1 ]
[ 3, 3, 0, 2, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
gsm8k/test/000030
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
openai/gsm8k
easy
math
true
[ 1, 1, 1, 0.6666666666666666, 1 ]
[ 3, 3, 3, 2, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
gsm8k/test/000048
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
openai/gsm8k
easy
math
true
[ 1, 1, 0.3333333333333333, 0, 1 ]
[ 3, 3, 1, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
gsm8k/test/000051
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
openai/gsm8k
easy
math
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
gsm8k/test/000103
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
hard
math
true
[ 1, 1, 1, 0, 1 ]
[ 3, 3, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000024
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
medium
math
true
[ 0.3333333333333333, 0.3333333333333333, 1, 0, 1 ]
[ 1, 1, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
math500/test/000037
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
medium
math
true
[ 1, 1, 1, 0.3333333333333333, 1 ]
[ 3, 3, 3, 1, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000061
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
hard
math
true
[ 1, 1, 1, 0, 1 ]
[ 3, 3, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000068
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
hard
math
true
[ 1, 1, 1, 0, 1 ]
[ 3, 3, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000092
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
easy
math
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000121
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
hard
math
true
[ 1, 1, 0.6666666666666666, 0, 1 ]
[ 3, 3, 2, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
math500/test/000150
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
easy
math
true
[ 1, 1, 1, 0.3333333333333333, 1 ]
[ 3, 3, 3, 1, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000160
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a" ]
HuggingFaceH4/MATH-500
hard
math
false
[ 1, 0.6666666666666666, 0, 0, 0 ]
[ 3, 2, 0, 0, 0 ]
[ 0, 1, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
math500/test/000168
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
easy
math
true
[ 1, 1, 1, 0, 1 ]
[ 3, 3, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
math500/test/000169
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
easy
math
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
math500/test/000178
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
medium
math
true
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
math500/test/000198
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
HuggingFaceH4/MATH-500
easy
math
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000207
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_b" ]
HuggingFaceH4/MATH-500
hard
math
false
[ 0, 1, 0, 0, 0.3333333333333333 ]
[ 0, 3, 0, 0, 1 ]
[ 0, 0, 0, 0, 2 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
math500/test/000219
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000012
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
mmlu/test/000021
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 1, 0, 1 ]
[ 3, 3, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000025
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 0.6666666666666666, 0, 0, 1 ]
[ 3, 2, 0, 0, 3 ]
[ 0, 1, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000037
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000046
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000072
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 0.3333333333333333, 0, 1 ]
[ 3, 3, 1, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
mmlu/test/000098
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000139
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu/test/000167
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_cheap_reasoning" ]
cais/mmlu
medium
knowledge
true
[ 1, 1, 1, 0, 1 ]
[ 3, 3, 3, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
test
in_distribution
mmlu/test/000174
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_mid", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 0, 1, 1 ]
[ 3, 3, 0, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu_pro/test/000053
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
mmlu_pro/test/000055
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 1, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu_pro/test/000070
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
mmlu_pro/test/000079
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 1, 1, 0.6666666666666666 ]
[ 3, 3, 3, 3, 2 ]
[ 0, 0, 0, 0, 1 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu_pro/test/000098
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 1, 1, 0.3333333333333333 ]
[ 3, 3, 3, 3, 1 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu_pro/test/000125
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu_pro/test/000146
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
mmlu_pro/test/000160
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 0, 0, 1 ]
[ 3, 3, 0, 0, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
train
in_distribution
mmlu_pro/test/000175
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
[ "frontier_a", "frontier_b", "frontier_c", "open_mid", "open_cheap_reasoning" ]
TIGER-Lab/MMLU-Pro
hard
knowledge
true
[ 1, 1, 1, 1, 1 ]
[ 3, 3, 3, 3, 3 ]
[ 0, 0, 0, 0, 0 ]
[ 0, 0, 0, 0, 0 ]
v1
3
validation
in_distribution
mmlu_pro/test/000193

BenchGen Router Pilot: per-question agent rewards

A routing dataset: for each question, how every agent in a fixed pool actually performed. It is the input a model-selection policy trains on — not a question-answering dataset.

Each row is one task. mean_reward[i] is the accuracy of agent agent_order[i] over 3 independent attempts. Ties are preserved explicitly rather than collapsed by argmax, because on easy questions several agents are genuinely equal and pretending otherwise teaches a router a coin flip.

Contents

46 tasks x 5 agents x 3 repetitions = 690 graded calls.

Facet Breakdown
Source HuggingFaceH4/MATH-500 14, TIGER-Lab/MMLU-Pro 10, allenai/ai2_arc 6, cais/mmlu 10, openai/gsm8k 4, opencompass/AIME2025 2
Domain knowledge 20, math 20, reasoning 6
Difficulty easy 9, hard 18, medium 19

The pool

# Agent Model Mode $/1M out
0 frontier_a openai/gpt-oss-120b reasoning 0.17
1 frontier_b deepseek/deepseek-v4-flash-0731 reasoning 0.18
2 frontier_c google/gemma-3-27b-it direct 0.45
4 open_mid mistralai/mistral-nemo direct 0.03
7 open_cheap_reasoning inclusionai/ling-3.0-flash reasoning 0.063

Collected but excluded

These were collected and are not in this release, because a label we cannot trust is worse than no label.

Source Why
rlpr Answers are free-form prose and tables. Our string-match grader scores correct answers as wrong, so these labels are not trustworthy.

What is deliberately not here

  • No question text. Upstream licences differ per source: ARC-Challenge is share-alike, and MMLU-Pro's licence was never verified. task_id identifies the upstream row so prompts can be rehydrated from the original datasets.
  • No raw completions. Provider terms on republishing model outputs were not cleared. The routing signal — correctness, ties, empties, errors — survives without them.

Limitations, stated plainly

  • High tie rate (43/46 tasks). Most agents solve most of these questions, so the per-question best agent is often arbitrary. A router trained on this alone learns little; the dataset is most useful as evidence that routing does not pay on saturated benchmarks.
  • One agent dominates. frontier_a or frontier_b is the best agent on almost every dataset here, so the pool behaves as a ranking rather than a set of specialists. That is a property of the pool, and no amount of extra data changes it.
  • The pool was hand-picked, not selected by measurement. Trinity's Appendix A.6 chooses agents and datasets jointly from a full screening matrix; that screening was skipped for cost.
  • No direct-vs-reasoning contrast. The intended pair of one base model in both modes was dropped: every candidate -thinking slug ignored max_tokens (24k-75k tokens against a 4,096 cap), which breaks protocol comparability.
  • No code domain. Execution-based grading needs a sandbox that does not exist here.
  • Small. This is a pilot, sized to a strict budget.

Protocol

Identical for every agent, because E(D, M) is not comparable otherwise: max_tokens 4096, temperature 0.1, top_p 1.0, minimal reasoning effort, 3 repetitions. Any response exceeding the token cap aborts collection rather than being recorded, so no row mixes protocols.

Grading

Answers are extracted by tagged answer, boxed LaTeX, labelled answer, then a last-number fallback that strips markup first. An empty response is never scored correct.

Downloads last month
28

Models trained or fine-tuned on benchgen/router-pilot