openai-community/gpt2

GPT-2 is a historically important, text-only autoregressive checkpoint rather than a competitive general-purpose model. Its Hugging Face entry supports standard Transformers deployment and the supplied repository establishes provenance, but neither provides contemporary capability measurements. It is substantially below Pegasus 1.5 even on text reasoning and far below GPT-5.6 Luna, Gemini 3.5 Flash, Kimi K2.5, GPT-5.5, and Claude Opus 4.8 in technical capability, coding, reasoning, and I/O. Unlike Pegasus, it has no multimodal modality evidence. For practical experimentation, local deployment avoids per-token API charges and permits broad runtime choice, supporting cost-effectiveness and availability where suitable hardware is available. The Transformers ecosystem and longstanding research use make the developer experience materially stronger than its task capability alone suggests. However, no supplied source states a license or a current hosted pricing schedule; cost is deployment-dependent, so pricing clarity is low. Its historical adoption signal is stronger than newer low-signal catalog models, but is not evidence of present-day preference or production fitness. No model-specific DeepSWE, LiveCodeBench, or LMArena result was supplied; the latter two checks explicitly found no listing. SWE-bench, Artificial Analysis, Terminal-Bench/Aider, and official documentation were checked, but the supplied evidence provides no scored GPT-2 result in those families. Consequently, the low coding and reasoning scores are conservative editorial placement based on the model's historical positioning, not inferred benchmark outcomes. Evidence quality is limited by archived and mismatched general OpenAI documentation, despite clear checkpoint provenance.