List your company

Long-form Writing Evaluation on WritingBench In-domain (test)

1,574Elo RatingGPT-5.4
769.04978.021,1871,395.98Jun 2, 2026

Evaluation results

MethodLinks
GPT-5.4Judge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,574
Claude Opus 4.6Judge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,468
HumanJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,464
POLARIS-9BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,419
Gemini 3.1 ProJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,387
Gemini 3 FlashJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,335
Qwen3.5-27BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,285
Gemma-4-31BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,283
Gemma-4-26B-A4B*Architecture=MoE, Judge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,235
Gemma-4-E4B†Architecture=PLE, Judge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,108
Qwen3.5-9BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,086
Ministral-3-8BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.061,055
DeepSeek-R1-Distill-Qwen-14BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.06891
LongWriter-Llama3.1-8BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.06868
LongWriter-Zero-32BJudge=Flash judge, Protocol=dual-position comparisons, Dimensions=11 judge dims + length, Number of prompts=45 ID prompts2026.06800