Hugging Face trending models: Week 27 snapshot

This first radar pass uses Hugging Face's public trending models page captured on June 30, 2026. The five cards follow the top five model entries shown there, so this is a snapshot of current platform momentum rather than a benchmark leaderboard. 1

Baidu's Unlimited-OCR is a 3B image-text model aimed at long-document OCR and document parsing; its model card describes the release as "one-shot long-horizon parsing" and lists 362,945 downloads in the last month. Recent release notes also mention vLLM inference support. 2

Qwythos-9B-Claude-Mythos-5-1M-GGUF packages a 9B reasoning model for llama.cpp, Ollama, LM Studio, Jan, and other GGUF runtimes. The card claims a 1,048,576-token context window and shows 907,682 monthly downloads, which explains why a small local-friendly build landed high in the trending list. 3

GLM-5.2 is Z.ai's 753B-parameter flagship model for long-horizon tasks. Its model card emphasizes a 1M-token context window, coding improvements, an MIT license, and local serving paths through SGLang, vLLM, Transformers, KTransformers, and Unsloth; Hugging Face lists 133,350 monthly downloads. 4

Ornith-1.0-35B-GGUF is a 35B GGUF model for agentic coding. The model card positions it as the lightweight member of the Ornith family, cites SWE-bench Verified at 75.6 for the 35B model, uses an MIT license, and lists 123,598 monthly downloads. 5

Qwen-AgentWorld-35B-A3B is a language world model for simulating agent environments. Its card says the model covers seven domains, including MCP tool calling, search, terminal, software engineering, Android, web, and OS environments; it also lists 35B total parameters, 3B activated parameters, a 262,144-token context length, and 26,223 monthly downloads. 6

Use the images as a fast triage pass. The monthly download numbers and rank position indicate attention on Hugging Face this week; they do not prove model quality. For production decisions, open the linked model card and check license, inference stack, evaluation setup, and known limitations before testing.