DeepSeek V3.2 Chat: benchmark results

Provider: DeepSeek. Released 2025-12-01. Access: Open.

Unified ELO 1741 ± 30, rank #256 of 2656 rated models, from 22 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
MCPMark29.72Pass@1 (%)68.4
AGI-Eval Community - Subject Knowledge85.33Accuracy (%)67.9
AGI-Eval Community - General Reasoning96.82Accuracy (%)65.7
AGI-Eval Community - Subject Reasoning (Chinese)85.29Accuracy (%)60.5
AGI-Eval Community - Subject Reasoning (English)83Accuracy (%)58
AGI-Eval Community - Subject Reasoning83.7Accuracy (%)57.5
AGI-Eval Community - Cognition (Chinese)76.52Accuracy (%)53.6
AGI-Eval Community - Learning (Chinese)78.16Accuracy (%)52.9
AGI-Eval Community - Interaction (English)90.06Accuracy (%)50
AGI-Eval Community - Interaction (Chinese)74.47Accuracy (%)44.9
AGI-Eval Community - Objective Accuracy (Chinese)57.38Accuracy (%)44.2
AGI-Eval Community - Objective Accuracy63.16Accuracy (%)42.9

Interactive version: theaggregate.ai/model?slug=deepseek-v3-2-chat · How It Works · Data refreshed daily, snapshot 2026-09-19.