Leaderboard / deepseek/deepseek-v4-pro
deepseek-v4-pro
deepseek · Tasks passed 89.0% · Finance Index 89.4
Tasks passed
89.0%
Overall accuracy
88.1%
Finance Index
89.4
Consistency
2.6/3 avg
Est. eval cost
$0.07
Median latency
15.6s
Median output speed
2 tok/s
- Harness version
- 0.2.0
- Task set
- v2
- Runs per task
- 3
- Evaluated
- 7/13/2026, 1:08:15 PM
- Run ID
- 75d557a3-fd82-4cc0-960e-a57bb260a786
- Results file
- Download JSON
Pass@1 by Difficulty
Easy
—
Medium
—
Hard
—
Scores by domain
Tasks passed (%) · top 1 models
Quant
Per-attempt drill-down requires a local result JSON in `results/`. Aggregate scores are shown from the published leaderboard.