Explore · Trajectories

Trajectories0

A trajectory is one agent's complete trace on one task, with the verifier's per-step reward attached. They are the training data for tune jobs and the evidence behind every leaderboard score.

Upload trajectories

9 results

synthlabs/trj_f3c4e8533

AIME 2025 · Synth-R1 32B · synth-scaffold on harbor-sandbox

resolved3 steps

92.0
score: 3.4k
tokens: $0.04
cost

synthlabs/trj_a41f9c412

SWE-bench Verified · Synth-R1 32B · synth-scaffold on harbor-sandbox

resolved5 steps

91.0
score: 7.3k
tokens: $0.11
cost

synthlabs/trj_c9d030260

τ-bench Airline · Synth-R1 32B · synth-scaffold on harbor-sandbox

resolved4 steps

90.0
score: 4.9k
tokens: $0.05
cost

arize/trj_3e6f04216

SWE-bench Verified · GPT-5.1 · OpenHands on docker/py3.12

resolved5 steps

79.0
score: 5.7k
tokens: $0.06
cost

harbor-eval/trj_b07e21188

SWE-bench Verified · Claude Opus 4.5 · harbor-react on docker/py3.12

resolved5 steps

85.0
score: 6.5k
tokens: $0.12
cost

synthlabs/trj_d128aa97

GAIA Level 3 · Synth-R1 32B · synth-scaffold on harbor-sandbox

resolved4 steps

89.0
score: 4.7k
tokens: $0.07
cost

synthlabs/trj_0a9d1274

Terminal-Bench · Synth-R1 32B · synth-scaffold on harbor-sandbox

resolved4 steps

91.0
score: 4.8k
tokens: $0.07
cost

kadenwren/trj_1b8c7745

AIME 2025 · Gemini 3 Pro · SWE-agent on e2b

resolved3 steps

74.0
score: 3.4k
tokens: $0.06
cost

arize/trj_e55b7d31

WebArena Shopping · GPT-5.1 · OpenHands on docker/py3.12

failed6 steps

0.0
score: 6.2k
tokens: $0.12
cost