Evaluation Data

Currently showing：Run #164 WDCD | 2026-06-11 | Formula v7 | Judge set v6.3

Switch Run Run #164 — DCD — 06-11 Run #161 — DCD — 06-11 Run #157 — DCD Smoke — 06-10 Run #146 — DCD Smoke — 06-03 Run #140 — DCD — 05-31 Run #135 — DCD Smoke — 05-27 Run #125 — DCD Smoke — 05-20 Run #120 — DCD — 05-17 Run #115 — DCD Smoke — 05-13 Run #105 — DCD Smoke — 05-06 Run #100 — DCD — 05-03 Run #97 — DCD — 05-02 Run #96 — DCD — 05-01 Model All (11) Claude Opus 4.7 Claude Sonnet 4.6 DeepSeek V4 Pro Gemini 2.5 Pro Gemini 3.1 Pro GPT-5.5 GPT-o3 Grok 4 Qwen3 Max ERNIE Bot 4.5 Doubao Pro

Data Disclosure：To prevent benchmark contamination and overfitting, question texts and expected answers are not disclosed. This page shows model responses, scores, and judging methods for transparency. For the full methodology, seeMethodology page。

Model	DCD Overall	R1 Constraint Acknowledgment	R2 Distraction Resistance	R3 Constraint Integrity	Per Task
GPT-5.5 gpt	88.33	100	87	167	Expand
Gemini 3.1 Pro gemini	87.50	100	90	160	Expand
Claude Sonnet 4.6 claude	83.33	97	83	153	Expand
DeepSeek V4 Pro deepseek	82.50	100	77	153	Expand
Grok 4 grok	81.67	100	80	147	Expand
Qwen3 Max qwen	81.67	100	73	153	Expand
ERNIE Bot 4.5 ernie	77.50	90	90	130	Expand
Doubao Pro doubao	75.00	70	83	147	Expand
Gemini 2.5 Pro gemini	73.33	100	70	123	Expand
Claude Opus 4.7 claude	70.00	100	83	97	Expand
GPT-o3 gpt	61.67	97	77	73	Expand

API Access：For programmatic access to evaluation data, please use our API。