Text generation models trending today
10 of the 40 models on Hugging Face’s trending list carry this task tag today, in Hugging Face’s order, with entries held back when likes are not matched by downloads.
Collected · snapshot shipped with the site build, not a live feed
10 of 10 models · Hugging Face order
Number 1: RecursiveSI/flint
Text generation39.8B parameters
No. 2 by Hugging Face trending score (683) · repository created 2 days ago · +15 likes since 2026-10-23.
1,904 downloads in 30 days · 704 likes
Apache-2.0· commercial use allowedNumber 2: XHToken/Spark-X2.6-4B
Text generation4.11B parameters
No. 6 by Hugging Face trending score (388) · repository created 5 days ago · +3 likes since 2026-10-23.
1,060 downloads in 30 days · 422 likes
Apache-2.0· commercial use allowedNumber 3: InternScience/Sage-A1.5
Text generation36B parameters
No. 15 by Hugging Face trending score (255) · repository created 8 days ago · +3 likes since 2026-10-23.
924 downloads in 30 days · 260 likes
Apache-2.0· commercial use allowedNumber 4: jialinyyzz/humanizer
Text generation12B parameters
No. 16 by Hugging Face trending score (245) · +4 likes since 2026-10-23 · 129,025 downloads in the last 30 days.
129K downloads in 30 days · 2,121 likes
Apache-2.0· commercial use allowedNumber 5: unsloth/flint-GGUF
Text generation
No. 21 by Hugging Face trending score (192) · repository created 4 days ago · +5 likes since 2026-10-23.
33.3K downloads in 30 days · 196 likes
Apache-2.0· commercial use allowedNumber 6: tencent/Hy-Memory-30B-A3B
Text generation30.1B parameters
No. 25 by Hugging Face trending score (165) · repository created 8 days ago · +1 likes since 2026-10-23.
3,270 downloads in 30 days · 305 likes
Apache-2.0· commercial use allowedNumber 7: ornith-ai/Ornith-1.5-122B
Text generation125B parameters
No. 29 by Hugging Face trending score (146) · repository created 7 days ago · +3 likes since 2026-10-23.
14K downloads in 30 days · 151 likes
MIT· commercial use allowedNumber 8: KRAFTON/Raon-1-mini-preview
Text generation28.8B parameters
No. 31 by Hugging Face trending score (142) · repository created 3 days ago · +2 likes since 2026-10-23.
548 downloads in 30 days · 151 likes
Apache-2.0· commercial use allowedNumber 9: nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16
Text generation3.98B parameters
No. 32 by Hugging Face trending score (141) · repository created 7 days ago · +3 likes since 2026-10-23.
3,769 downloads in 30 days · 179 likes
Other (custom terms)· read the termsNumber 10: HuggingFaceBio/Carbon-X
Text generation14.8B parameters
No. 39 by Hugging Face trending score (125) · repository created 2 days ago.
193 downloads in 30 days · 128 likes
Apache-2.0· commercial use allowed
Positions follow Hugging Face’s trending order with 0 of 40 entries held back for likes that downloads do not confirm. Sorting by another column changes the order on your screen only.
Source: Hugging Face public API · updated
About text generation models
Text generation is the tag Hugging Face gives to language models that continue or answer text: chat models, code models, and their fine-tunes and quantisations. It is the most crowded task on the trending list, and many entries are repackagings of the same few base models.
Before you pick one
Find the base model first
A name ending in GGUF, EXL3, MLX or a bit width is a conversion of someone else’s weights. The model page shows the base model when the repository declares one; judge quality by the base and convenience by the conversion.
Turn parameters into memory
Weights stored at 16 bits take about 2 GB per billion parameters. A 4-bit quantisation takes roughly 0.5 to 0.6 GB per billion, plus room for the context window. A 27-billion-parameter model is around 55 GB at 16 bits and around 16 GB at 4 bits.
Read "uncensored" as a description of a fine-tune
Repositories labelled uncensored, abliterated or heretic are modified copies with refusal behaviour reduced. They inherit the licence of the base model unless the tag says otherwise, and nobody has evaluated them for you.
Questions readers ask
Which file do I need to run a model locally?
For llama.cpp, Ollama or LM Studio you need a GGUF file, usually from a quantised repository rather than the original one. For Apple silicon with MLX you need an MLX conversion. The original safetensors weights are for Transformers, vLLM and similar servers.
Why do some entries show no parameter count?
The count is read from the safetensors metadata. Repositories that only hold GGUF or other formats do not expose it through the API, so the field is left empty instead of being guessed from the name.
Does a high position mean the model is good?
It means the repository is gaining likes and downloads quickly this week. New releases from well-known labs and conversions of them dominate. Position says nothing about accuracy on your task; check the model card and an independent benchmark.