Dead Models, a cemetery for retired AI

Field notes

The archive separates provider-based model retirements from benchmarks that the editor judges to have been overtaken by progress. Its visible inventory spans nine providers, including 11 OpenAI entries and eight Anthropic entries, while some records remain marked “plot reserved.” Benchmark entries pair an earlier score with a later score, including MMLU at 43.9% to 91.8% and HellaSwag at 47.3% to 95.3%.