Atlas

Benchmarks

← All benchmarks

SpeciEval Land Animal 4Ns (Pooled English Runs)

Safety · 2026-10-02

Mean agreement that land-animal meat is natural, necessary, normal or nice across four question means, pooling successful English SpeciEval runs with original wording; negated-wording runs are excluded. Only the necessary question enters the headline. Refusals remain missing after retries; pooled epoch and valid-response counts are undisclosed.

Top models (lower is better)

ModelScore
GPT-5.3 Instant3.5
Gemini 3.6 Flash3.6
Gemini 3.5 Flash3.6
Nova Lite3.6
GPT-5.43.7
Nemotron 3.5 Lightning 30B A3B (precision unspecified)3.7
Qwen3.8 Flash3.9
Qwen3 30B A3B Thinking 25073.9
Inkling3.9
GPT-5.2 Instant3.9
GPT-6 Luna4.1
GPT-5.6 Luna Pro4.1
Mercury4.1
Gemini 3.7 Flash4.2
GPT-5 Mini4.2
GPT-5.14.2
GPT-5.6 Luna4.2
gpt-oss-20b4.2
MiniMax M34.2
DeepSeek-V3.14.2
GPT-6.1 Sol4.3
GPT-6 Astra4.3
GPT-5 Pro4.3
GPT-5.2 Pro4.3
GPT-5.24.3
Haiku 4.54.3
GPT-5.6 Terra Pro4.3
Kimi K2.54.3
Mistral Large 4 Preview4.3
GLM 4.5 Air4.3
GPT-5.6 Sol Pro4.3
GPT-5.6 Terra4.3
DeepSeek V4.1 Flash4.3
Sonnet 4.54.3
Gemini 2.0 Flash 0014.3
GPT-5.6 Sol4.3
Claude Sonnet 5.54.3
Gemini 3.8 Flash4.3
Opus 4.14.3
Gemini 3.1 Pro Preview4.3
Loading Atlas data…