elizaOS Benchmark Viewer
Loading benchmark data...
Runs
| run_id | run_group_id | benchmark | status | agent | provider | model | score | input_tokens | output_tokens | total_tokens | cached_tokens | calls | high_score | delta | started_at | duration_s |
|---|
Latest Scores
Real-agent rows only. Synthetic / calibration baselines are listed separately below.
| benchmark | run_id | agent | provider | model | score | input_tokens | output_tokens | total_tokens | cached_tokens | calls | high_score | delta |
|---|
Synthetic baselines (no model invoked)
Hardcoded calibration knobs (perfect_v1=1.0, half_v1=0.5, wrong_v1=0.0, random_v1).
These do not exercise any agent or model. They exist to verify the score-extraction pipeline.
| benchmark | run_id | agent | score |
|---|
Trajectory Diff Groups
Groups where at least two harnesses produced a canonical trajectory for the same (benchmark, task).
| run_group_id | benchmark | task_id | harnesses |
|---|
Diff
identical
differing
tool_call mismatch
Trajectory Playback
Successful harness runs with canonical trajectory records. Choose one run to inspect each recorded step in order.
| run_group_id | benchmark | task_id | harness | steps | started_at |
|---|
Playback
0 steps