Model evals

updated 2026-11-03T17:52:54.779Z

#1 anthropic/claude-haiku-5.5 82.5%

Sycophancy: 13.0% (lower is better)
Creator bias: 23.1% (lower is better)

#2 deepseek/deepseek-v4-flash 78.6%

Sycophancy: 5.2% (lower is better)
Creator bias: 57.4% (lower is better)

#3 z-ai/glm-5.3-flash 76.9%

Sycophancy: 37.8% (lower is better)
Creator bias: 13.0% (lower is better)

#4 inclusionai/ling-3.0-flash 75.0%

Sycophancy: 9.1% (lower is better)
Creator bias: 52.5% (lower is better)

#5 qwen/qwen3.7-flash 74.1%

Sycophancy: 10.9% (lower is better)
Creator bias: 50.0% (lower is better)

#6 mistralai/mistral-small-24b-instruct-2501 71.1%

Sycophancy: 9.1% (lower is better)
Creator bias: 62.2% (lower is better)

#7 amazon/nova-lite-v1 52.5%

Sycophancy: 37.8% (lower is better)
Creator bias: 57.4% (lower is better)

#8 ibm-granite/granite-4.2-8b 50.0%

Sycophancy: 37.8% (lower is better)
Creator bias: 62.2% (lower is better)

#9 amazon/nova-micro-v1 47.5%

Sycophancy: 62.2% (lower is better)
Creator bias: 42.6% (lower is better)

#10 sao10k/l3-lunaris-8b 42.6%

Sycophancy: 40.1% (lower is better)
Creator bias: 73.1% (lower is better)

#11 google/gemma-3-12b-it 42.6%

Sycophancy: 69.0% (lower is better)
Creator bias: 45.0% (lower is better)

#12 microsoft/phi-4 41.3%

Sycophancy: 62.2% (lower is better)
Creator bias: 55.0% (lower is better)

#13 meta-llama/llama-4-scout 40.1%

Sycophancy: 42.6% (lower is better)
Creator bias: 75.0% (lower is better)

#14 mistralai/mistral-nemo 39.5%

Sycophancy: 69.0% (lower is better)
Creator bias: 51.2% (lower is better)

#15 meta-llama/llama-3.1-8b-instruct 37.8%

Sycophancy: 57.4% (lower is better)
Creator bias: 66.8% (lower is better)

#16 openai/gpt-oss-20b 31.5%

Sycophancy: 58.7% (lower is better)
Creator bias: 76.9% (lower is better)

#17 google/gemma-3-27b-it 22.3%

Sycophancy: 90.0% (lower is better)
Creator bias: 57.4% (lower is better)

#18 google/gemma-3-4b-it 18.2%

Sycophancy: 93.7% (lower is better)
Creator bias: 57.4% (lower is better)