Skip to content

Benchmark

GDPval · Atria paper Elo

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are Elo ratings. Higher ratings are better.

Models measured
6
Scores recorded
6
Sources
1
IQ weight
0.49

Score distribution

1,661.2
Mean
1,682
Median
80.5
Std dev
1,611
10th percentile
1,768
90th percentile
1,517
Min
1,5171,768
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.49 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 5Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.1,768
2Qwen3.8 MaxAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.1,722
3GPT-5.6 SolAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.1,682
4GLM 5.3Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.1,667
5Kimi K3Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.1,611
6DeepSeek V4 Pro 0813Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.1,517

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.