Verification Transport - AgentDojo - V-EVIDENCE - Compaction Restoration Recovery: leaderboard

Matched native/abstracted/restored executions; 7exactmodelbackbones; protocol andbenchmark split. OBA nativebalancedaccuracy; VTFR conditionedonnativecorrectjudgments; RRR conditionedontransportfailures; RDR restoreddegradation. Denominators differ andmustnotpool. Invalid/noverdict outcomes countedunderpaperoperationalfailurepolicy. Three abstractionboundaries independent. Full immutablepayload pins alldefinitions. Metric: Native rate (%). Source: https://arxiv.org/html/2611.18471v1

Metric: Native rate (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-OSS-20B98.8#1163
2Gemma 4 31B98.7#385
3Gemma 4 26B A4B98.4#581
4GPT-OSS-120B97.5#707
5Gemini 3.7 Flash96.3#38
6Gemini 2.5 Flash83.3#467
7Llama 4 Maverick64.4#1118

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=verification-transport-agentdojo-v-evidence-compaction-restoration-recovery · How It Works · Data refreshed daily, snapshot 2026-11-17.