Datasets:

moncefem
/

featherless-ai-models

id large_stringlengths 16 16	model_name large_stringlengths 6 84	organisation_name large_stringlengths 3 20	parameter_count_raw large_stringclasses 42 values	context_length_raw large_stringclasses 5 values	model_type_raw large_stringlengths 6 19 ⌀	download_count_raw large_stringlengths 1 5	published_date large_stringdate 2023-07-01 00:00:00 2026-05-01 00:00:00 ⌀	n_parameters_billions float64 0.5 1k	n_context_length_tokens int64 2k 32k	n_downloads_absolute int64 0 18.5M	cat_modality large_stringclasses 4 values	is_instruction_tuned bool 2 classes	is_quantized bool 2 classes	has_vision_capability bool 2 classes	cat_org_category large_stringclasses 5 values	dn_year float64 2.02k 2.03k ⌀	dn_month float64 1 12 ⌀	is_high_popularity bool 1 class	is_large_model bool 2 classes	is_long_context bool 1 class
sig_de859b993a4c	DeepSeek-V4-Pro	deepseek-ai	862B	32K	deepseek4-1.6t	3.6M	2026-04-01	862	32,000	3,600,000	text	null	null	null	research_lab	2,026	4	null	true	false
sig_87640113c9d4	DeepSeek-V4-Flash	deepseek-ai	284B	32K	deepseek4-284b	2.0M	2026-04-01	284	32,000	2,000,000	text	null	null	null	research_lab	2,026	4	null	true	false
sig_b35472da18cf	Qwen3.6-35B-A3B	Qwen	35B	32K	Vision	5.7M	2026-04-01	35	32,000	5,700,000	vision	null	null	true	null	2,026	4	null	false	false
sig_2a7c15bac634	Qwen3.6-27B	Qwen	28B	32K	Vision	3.7M	2026-04-01	28	32,000	3,700,000	vision	null	null	true	null	2,026	4	null	false	false
sig_bff8f953c4af	gemma-4-31B-it	google	31B	32K	gemma4-31b	10.0M	2026-03-01	31	32,000	10,000,000	text	true	false	false	big_tech	2,026	3	null	false	false
sig_b491528ce513	gemma-4-E4B-it	google	4B	32K	gemma4-e4b	6.2M	2026-03-01	4	32,000	6,200,000	text	true	false	false	big_tech	2,026	3	null	false	false
sig_bc77f1144758	GLM-5.1	zai-org	754B	32K	glm51-754b	244k	2026-04-01	754	32,000	244,000	null	null	null	null	null	2,026	4	null	true	false
sig_ef5d8309e291	Llama-3.1-8B-Instruct	meta-llama	8B	32K	llama31-8b	10.8M	2024-07-01	8	32,000	10,800,000	text	true	false	false	big_tech	2,024	7	null	false	false
sig_76545eae6154	gemma-4-26B-A4B-it	google	26B	32K	gemma4-26b	9.0M	2026-03-01	26	32,000	9,000,000	text	true	true	false	big_tech	2,026	3	null	false	false
sig_936476ff90f3	Kimi-K2.6	moonshotai	1000B	32K	Vision	2.5M	2026-04-01	1,000	32,000	2,500,000	vision	null	null	true	startup	2,026	4	null	true	false
sig_4516d59da9fe	Qwen3.5-9B	Qwen	9B	32K	Vision	8.0M	2026-02-01	9	32,000	8,000,000	vision	null	null	true	null	2,026	2	null	false	false
sig_e4677081a840	gemma-4-E2B-it	google	2B	32K	gemma4-e2b	3.4M	2026-03-01	2	32,000	3,400,000	text	true	null	false	big_tech	2,026	3	null	false	false
sig_32d625aa434e	WebWorld-8B	Qwen	8B	32K	qwen3-8b	2k	2026-02-01	8	32,000	2,000	text	false	false	false	research_lab	2,026	2	false	false	false
sig_46cba097ba58	Qwen3.5-4B	Qwen	4B	32K	Vision	7.9M	2026-02-01	4	32,000	7,900,000	vision	null	null	true	null	2,026	2	null	false	false
sig_d78c08ccd0df	gpt-oss-20b	openai	20B	32K	gpt-oss-20b	7.7M	2025-08-01	20	32,000	7,700,000	text	false	false	false	big_tech	2,025	8	null	false	false
sig_fd8a8e216334	gpt-oss-120b	openai	120B	32K	gpt-oss-120b	4.8M	2025-08-01	120	32,000	4,800,000	text	null	null	null	big_tech	2,025	8	null	true	false
sig_6fbe523d97a9	MiniMax-M2.7	MiniMaxAI	229B	32K	minimax-m2-228b7	548k	2026-04-01	229	32,000	548,000	null	null	null	null	null	2,026	4	null	true	false
sig_f6d687a76c1e	gemma-4-E4B	google	4B	32K	gemma4-e4b	1.3M	2026-03-01	4	32,000	1,300,000	text	false	false	false	big_tech	2,026	3	null	false	false
sig_96b9976c9204	gemma-4-E2B	google	2B	32K	gemma4-e2b	1.0M	2026-03-01	2	32,000	1,000,000	text	false	false	false	big_tech	2,026	3	null	false	false
sig_b04330a96583	gemma-4-26B-A4B	google	26B	32K	gemma4-26b	222k	2026-03-01	26	32,000	222,000	text	null	null	null	big_tech	2,026	3	null	false	false
sig_dcd8e0e0be85	Qwen/Qwen2.5-7B-Instruct	Qwen	8B	32K	qwen25-7b	13.0M	2024-09-01	8	32,000	13,000,000	text	true	false	false	big_tech	2,024	9	null	false	false
sig_54335d74ab78	Qwen/Qwen3-1.7B	Qwen	2B	32K	qwen3-1b7	3.6M	2025-04-01	2	32,000	3,600,000	text	false	false	false	research_lab	2,025	4	null	false	false
sig_bd12f943c67a	Qwen/Qwen3-Embedding-0.6B	Qwen	600M	32K	qwen3-embedding-0b6	6.2M	2025-06-01	0.6	32,000	6,200,000	embedding	false	false	false	research_lab	2,025	6	null	false	false
sig_b69315257b7b	nvidia/Gemma-4-31B-IT-NVFP4	nvidia	31B	32K	gemma4-31b	2.2M	2026-04-01	31	32,000	2,200,000	text	true	true	false	big_tech	2,026	4	null	false	false
sig_44514347c8f7	Qwen/Qwen3-Coder-30B-A3B-Instruct	Qwen	30B	32K	qwen3moe-30b	1.8M	2025-07-01	30	32,000	1,800,000	text	true	false	false	research_lab	2,025	7	null	false	false
sig_733d273e69dd	moonshotai/Kimi-K2.5	moonshotai	1000B	32K	kimi-k25	1.7M	2026-01-01	1,000	32,000	1,700,000	null	null	null	null	null	2,026	1	null	true	false
sig_1b6322f27bfb	mistralai/Mistral-7B-Instruct-v0.2	mistralai	7B	8K	mistral-v02-7b	3.3M	2023-12-01	7	8,000	3,300,000	text	true	false	false	startup	2,023	12	null	false	false
sig_a0bd1a517004	NousResearch/Hermes-3-Llama-3.1-8B	NousResearch	8B	32K	llama31-8b	217k	2024-07-01	8	32,000	217,000	text	null	null	false	research_lab	2,024	7	null	false	false
sig_d5a83602a789	meta-llama/Llama-3.2-1B-Instruct	meta-llama	1B	32K	llama32-1b	7.7M	2024-09-01	1	32,000	7,700,000	text	true	null	null	big_tech	2,024	9	null	false	false
sig_fd1441408493	Qwen/Qwen3-0.6B	Qwen	800M	32K	qwen3-0b6	18.5M	2025-04-01	0.8	32,000	18,500,000	text	false	false	false	research_lab	2,025	4	null	false	false
sig_d28604b21edc	dphn/Dolphin-Mistral-24B-Venice-Edition	dphn	24B	32K	mistral-24b	13k	2025-06-01	24	32,000	13,000	text	null	null	false	null	2,025	6	null	false	false
sig_adadd36664d4	Qwen/Qwen3-Coder-Next	Qwen	80B	32K	qwen3moe-80b	1.1M	2026-01-01	80	32,000	1,100,000	text	null	null	false	research_lab	2,026	1	null	true	false
sig_b9a535541eb8	nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16	nvidia	120B	32K	nemotron3-120b	722k	2026-03-01	120	32,000	722,000	text	false	false	false	big_tech	2,026	3	null	true	false
sig_408c37c3f30a	meta-llama/Meta-Llama-3-8B-Instruct	meta-llama	8B	8K	llama3-8b	1.7M	2024-04-01	8	8,000	1,700,000	text	true	false	false	big_tech	2,024	4	null	false	false
sig_7b4244cdb4de	Qwen/Qwen2.5-0.5B-Instruct	Qwen	500M	32K	qwen25-0b5	4.9M	2024-09-01	0.5	32,000	4,900,000	text	true	false	false	research_lab	2,024	9	null	false	false
sig_68c4535f67bb	meta-llama/Llama-3.2-3B	meta-llama	3B	32K	llama32-3b	1.2M	2024-09-01	3	32,000	1,200,000	text	null	null	null	big_tech	2,024	9	null	false	false
sig_24356b96ec1a	Qwen/QwQ-32B	Qwen	33B	32K	qwen25-32b	62k	2025-03-01	33	32,000	62,000	text	null	null	false	big_tech	2,025	3	null	false	false
sig_5042cf7fdf23	Qwen/Qwen3-Embedding-8B	Qwen	8B	32K	qwen3-embedding-8b	1.6M	2025-06-01	8	32,000	1,600,000	embedding	false	false	false	null	2,025	6	null	false	false
sig_2834e1a54537	Qwen/Qwen3-VL-30B-A3B-Instruct	Qwen	30B	32K	qwen3vlmoe-30b	1.2M	2025-09-01	30	32,000	1,200,000	multimodal	true	null	true	research_lab	2,025	9	null	false	false
sig_3849e1061738	DavidAU/Qwen3-0.6B-heretic-abliterated-uncensored	DavidAU	800M	32K	qwen3-0b6	938	2025-11-01	0.8	32,000	938	text	null	null	false	community	2,025	11	false	false	false
sig_2f9e9456a82a	google/gemma-3-4b-it	google	4B	32K	Vision	2.1M	2025-02-01	4	32,000	2,100,000	vision	true	null	true	big_tech	2,025	2	null	false	false
sig_0af17a7e359e	google/gemma-3-12b-it	google	12B	32K	Vision	2.8M	2025-03-01	12	32,000	2,800,000	vision	true	false	true	big_tech	2,025	3	null	false	false
sig_cde255d96ee4	google/gemma-3-1b-it	google	1B	32K	null	902k	2025-03-01	1	32,000	902,000	text	true	null	null	big_tech	2,025	3	null	false	false
sig_30e51503d1c0	zemelee/qwen2.5-jailbreak	zemelee	3B	32K	null	182	2025-05-01	3	32,000	182	null	null	null	null	null	2,025	5	null	false	false
sig_4c38670ddd0b	Qwen/Qwen3-Embedding-4B	Qwen	4B	32K	Embedding	2.6M	2025-06-01	4	32,000	2,600,000	embedding	false	false	false	null	2,025	6	null	false	false
sig_8ab31bb80a05	Qwen/Qwen3-4B-Instruct-2507	Qwen	4B	32K	null	6.7M	2025-08-01	4	32,000	6,700,000	text	true	false	false	null	2,025	8	null	false	false
sig_6c057d889532	TheDrummer/Cydonia-24B-v4.3	TheDrummer	24B	32K	null	5k	2025-11-01	24	32,000	5,000	null	null	null	null	null	2,025	11	null	false	false
sig_06ad0ea50d58	zai-org/GLM-4.7-Flash	zai-org	30B	32K	null	780k	2026-01-01	30	32,000	780,000	null	null	null	null	null	2,026	1	null	false	false
sig_e5fb9eecad52	OBLITERATUS/Qwen3-4B-OBLITERATED	OBLITERATUS	4B	32K	null	247	2026-03-01	4	32,000	247	null	null	null	null	null	2,026	3	null	false	false
sig_889d0582fecf	TinyLlama/TinyLlama-1.1B-Chat-v1.0	TinyLlama	1B	2K	null	2.7M	2023-12-01	1.1	2,000	2,700,000	text	true	false	false	community	2,023	12	null	false	false
sig_8be3b6ae91d9	google/gemma-3-12b-it-qat-q4_0-unquantized	google	12B	32K	Vision	60k	2025-04-01	12	32,000	60,000	vision	true	true	true	big_tech	2,025	4	null	false	false
sig_a9e998fe4280	Qwen/Qwen3-4B	Qwen	4B	32K	null	7.2M	2025-04-01	4	32,000	7,200,000	null	null	null	null	null	2,025	4	null	false	false
sig_5cc595612a33	Qwen/Qwen3-14B	Qwen	14B	32K	null	2.4M	2025-04-01	14	32,000	2,400,000	null	null	null	null	null	2,025	4	null	false	false
sig_92066feb2d22	deepseek-ai/DeepSeek-R1-0528-Qwen3-8B	deepseek-ai	8B	32K	null	147k	2025-05-01	8	32,000	147,000	text	null	null	null	research_lab	2,025	5	null	false	false
sig_cebf93b4b52d	Tesslate/UIGEN-X-8B	Tesslate	8B	32K	null	26	2025-07-01	8	32,000	26	null	null	null	null	null	2,025	7	null	false	false
sig_72615ae53d44	swiss-ai/Apertus-8B-Instruct-2509	swiss-ai	8B	32K	null	119k	2025-08-01	8	32,000	119,000	text	true	false	false	null	2,025	8	null	false	false
sig_597ca9794a62	moonshotai/Kimi-K2-Instruct-0905	moonshotai	1000B	32K	null	1.8M	2025-09-01	1,000	32,000	1,800,000	text	true	false	false	startup	2,025	9	null	true	false
sig_10ac55df4568	fdtn-ai/Foundation-Sec-8B-Reasoning	fdtn-ai	8B	32K	null	12k	2025-11-01	8	32,000	12,000	null	null	null	null	null	2,025	11	null	false	false
sig_f26b5156aae2	deepseek-ai/DeepSeek-V3.2	deepseek-ai	685B	32K	null	4.3M	2025-12-01	685	32,000	4,300,000	null	null	null	null	null	2,025	12	null	true	false
sig_9c706fa87fee	Qwen/Qwen3.5-397B-A17B	Qwen	397B	32K	Vision	1.1M	2026-02-01	397	32,000	1,100,000	vision	null	null	true	null	2,026	2	null	true	false
sig_81b99cc4065d	deepseek-ai/DeepSeek-R1-Distill-Qwen-7B	deepseek-ai	8B	32K	qwen25-7b	615k	2025-01-01	8	32,000	615,000	text	null	null	false	research_lab	2,025	1	null	false	false
sig_0bb2b0f170c7	unsloth/Llama-3.1-8B-Instruct	unsloth	8B	32K	llama31-8b	418k	2025-02-01	8	32,000	418,000	text	true	null	false	community	2,025	2	null	false	false
sig_47749c906dbd	mistralai/Mistral-Small-3.1-24B-Instruct-2503	mistralai	24B	32K	Vision	396k	2025-03-01	24	32,000	396,000	vision	true	null	true	startup	2,025	3	null	false	false
sig_9257413c2ad7	mlabonne/gemma-3-27b-it-abliterated	mlabonne	27B	32K	Vision	325k	2025-03-01	27	32,000	325,000	vision	true	false	true	community	2,025	3	null	false	false
sig_455d97c96774	deepseek-ai/DeepSeek-V3-0324	deepseek-ai	685B	32K	deepseek-v3-lc	533k	2025-03-01	685	32,000	533,000	text	false	false	false	research_lab	2,025	3	null	true	false
sig_a14e6c31842c	IAAR-Shanghai/xVerify-0.5B-I	IAAR-Shanghai	500M	32K	qwen2-0b5	63	2025-03-01	0.5	32,000	63	text	true	null	false	research_lab	2,025	3	false	false	false
sig_43c25561e462	Qwen/Qwen3-8B-Base	Qwen	8B	32K	qwen3-8b	430k	2025-04-01	8	32,000	430,000	text	false	false	false	research_lab	2,025	4	null	false	false
sig_a3bf223da5e7	Qwen/Qwen3-4B-Base	Qwen	4B	32K	qwen3-4b	694k	2025-04-01	4	32,000	694,000	text	false	false	null	research_lab	2,025	4	null	false	false
sig_dbf6344c2f84	PocketDoc/Dans-PersonalityEngine-V1.3.0-24b	PocketDoc	24B	32K	mistral-24b	230	2025-05-01	24	32,000	230	text	null	null	false	null	2,025	5	null	false	false
sig_758722c5aa6c	katanemo/Arch-Router-1.5B	katanemo	2B	32K	qwen25-1b5	3k	2025-05-01	2	32,000	3,000	text	null	null	false	null	2,025	5	null	false	false
sig_6f89ec3524bd	Anubis-70B-v1.2	TheDrummer	70B	32K	llama31-70b	752	2026-01-01	70	32,000	752	text	null	null	false	community	2,026	1	null	true	false
sig_067110b7a9d8	Nanbeige4.1-3B	Nanbeige	4B	32K	nanbeige41-3b	205k	2026-02-01	4	32,000	205,000	text	null	null	null	null	2,026	2	null	false	false
sig_6b4544509af7	gemma-3-12b-it-heretic-v2	DreamFast	12B	32K	Vision	6k	2026-03-01	12	32,000	6,000	vision	true	null	true	startup	2,026	3	null	false	false
sig_1111b4d221b9	GRM-1.5b	OrionLLM	2B	32K	qwen2-1b5	384	2026-03-01	2	32,000	384	text	null	null	null	null	2,026	3	null	false	false
sig_e32d7212b204	sweep-next-edit-v2-7B	sweepai	8B	32K	qwen2-7b	1k	2026-03-01	8	32,000	1,000	text	null	null	null	startup	2,026	3	false	false	false
sig_b3a1adb4b268	Trinity-Large-Thinking	arcee-ai	399B	32K	afmoe-399b	21k	2026-04-01	399	32,000	21,000	text	null	null	null	startup	2,026	4	null	true	false
sig_0a839df86c2e	Qwen3-4B-Qwen3.6-plus-Reasoning-Slerp	khazarai	4B	32K	qwen3-4b	135	2026-04-01	4	32,000	135	text	null	null	false	community	2,026	4	false	false	false
sig_e9f17b0ba67d	llama3.1-heretic-unsensored	anasali4151	8B	8K	llama3-8b	172	2026-05-01	8	8,000	172	text	null	null	false	community	2,026	5	false	false	false
sig_70cedadf0dcd	Llama-3.1-8B	meta-llama	8B	32K	llama31-8b	1.4M	2024-07-01	8	32,000	1,400,000	text	false	false	false	big_tech	2,024	7	null	false	false
sig_6d5f4cf96828	Llama-3.3-70B-Instruct	meta-llama	70B	32K	llama33-70b	989k	2024-11-01	70	32,000	989,000	text	true	false	false	big_tech	2,024	11	null	true	false
sig_1199f9e2d8ac	Lily-Cybersecurity-7B-v0.2	segolilylabs	7B	4K	mistral-v01-7b	661	2024-01-01	7	4,000	661	text	false	false	false	community	2,024	1	null	false	false
sig_f000279a93a5	SecurityLLM	ZySec-AI	7B	4K	mistral-v01-7b	363	2024-01-01	7	4,000	363	text	null	null	false	startup	2,024	1	null	false	false
sig_9525308d5e25	Saul-7B-Instruct-v1	Equall	7B	4K	mistral-v01-7b	6k	2024-02-01	7	4,000	6,000	text	true	null	false	null	2,024	2	null	false	false
sig_3f54c884bd35	gemma-2b	google	3B	8K	gemma2-2b	275k	2024-02-01	3	8,000	275,000	text	false	false	false	big_tech	2,024	2	null	false	false
sig_9cbb6e481bb2	Gemma2_Farsi	alibidaran	3B	8K	gemma2-2b	10	2024-04-01	3	8,000	10	text	null	null	false	community	2,024	4	false	false	false
sig_edfc5b18c680	WizardLM-2-8x22B	alpindale	141B	32K	mixtral-8x22b	11k	2024-04-01	141	32,000	11,000	text	null	null	false	community	2,024	4	null	true	false
sig_fdef1146c699	dolphin-2.9-llama3-8b	dphn	8B	8K	llama3-8b	9k	2024-04-01	8	8,000	9,000	text	null	null	false	null	2,024	4	null	false	false
sig_772a63be55e9	Llama-3-8B-Lexi-Uncensored	Orenguteng	8B	8K	llama3-8b	1k	2024-04-01	8	8,000	1,000	text	false	false	false	community	2,024	4	false	false	false
sig_af8adc842b0c	Kocdigital-LLM-8b-v0.1	KOCDIGITAL	8B	8K	llama3-8b	12	2024-05-01	8	8,000	12	text	null	false	false	null	2,024	5	null	false	false
sig_8ccea56a1b44	llama-3-70B-Instruct-abliterated	failspy	70B	8K	llama3-70b	9k	2024-05-01	70	8,000	9,000	text	true	false	false	community	2,024	5	null	true	false
sig_148c85392452	meta-llama/Llama-3.1-70B	meta-llama	70B	32K	llama31-70b	105k	2024-07-01	70	32,000	105,000	text	null	null	false	big_tech	2,024	7	null	true	false
sig_9f3276b6a4c8	AI-MO/NuminaMath-72B-CoT	AI-MO	73B	32K	qwen2-72b	34	2024-07-01	73	32,000	34	text	false	false	false	unknown	2,024	7	false	true	false
sig_6cfbae6e408e	google/gemma-2-2b-it	google	3B	8K	gemma2-2b	387k	2024-07-01	3	8,000	387,000	text	true	null	false	big_tech	2,024	7	null	false	false
sig_1769e968118f	meta-llama/Llama-3.1-70B-Instruct	meta-llama	70B	32K	llama31-70b	775k	2024-07-01	70	32,000	775,000	text	true	false	false	big_tech	2,024	7	null	true	false
sig_e14f77724ead	mistralai/Mistral-Nemo-Base-2407	mistralai	12B	32K	mistral-nemo	56k	2024-07-01	12	32,000	56,000	text	false	false	false	startup	2,024	7	null	false	false
sig_983d3b34a213	mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated	mlabonne	8B	32K	llama31-8b	6k	2024-07-01	8	32,000	6,000	text	true	null	false	community	2,024	7	null	false	false
sig_2765e64d7cc7	TheDrummer/Gemmasutra-Mini-2B-v1	TheDrummer	3B	8K	gemma2-2b	182	2024-08-01	3	8,000	182	text	null	null	false	community	2,024	8	false	false	false
sig_115ccd175f24	betterdataai/PII_DETECTION_MODEL	betterdataai	500M	32K	qwen25-0b5	325	2024-08-01	0.5	32,000	325	text	null	null	false	null	2,024	8	null	false	false
sig_cb82ea4e13c2	Gryphe/Pantheon-RP-1.6-12b-Nemo	Gryphe	12B	32K	mistral-nemo	7	2024-08-01	12	32,000	7	text	false	false	false	community	2,024	8	false	false	false
sig_480938625633	MarinaraSpaghetti/NemoMix-Unleashed-12B	MarinaraSpaghetti	12B	32K	mistral-nemo	517	2024-08-01	12	32,000	517	text	null	null	false	community	2,024	8	null	false	false

End of preview. Expand in Data Studio

Featherless.ai Model Catalog — 300 Models, 20 Typed Variables

A typed, analysis-ready snapshot of 300 open-weights AI models listed on the first pages of Featherless.ai — an inference platform that hosts a long-tail of community and lab releases — joined with 20 LLM-extracted analytical variables (parameter count, context length, modality, organisation type, popularity tier, capability flags).

Generated end-to-end (scrape → typed schema → per-row LLM extraction → export) by Gemma Miner — an autonomous text-to-dataset agent that turns any website into a research-grade dataset in minutes.

TL;DR — what this dataset reveals about the open-weights ecosystem

One organisation (Qwen / Alibaba) commands 52 % of total downloads with 14 % of the models. 41 of the 300 listings carry the Qwen org tag; together they account for 118 M of the 227 M total downloads on the platform.
The "workhorse" tier is 4–10 B parameters. 110 of 300 models (37 %) sit in this bucket — bigger than every other size class combined, except the 10–30 B mid-tier.
Bigger ≠ more popular. The log–log Pearson correlation between parameter count and downloads is only r = 0.18. The most-downloaded model in the dataset is Qwen3-0.6B with 18.5 M downloads; some 200 B+ models have under 10 K.
Big-tech median popularity is ~3 000× community median. Median downloads: big-tech 945 500 · research-lab 252 500 · startup 2 000 · community 307. The platform's catalogue is a long tail dominated by community fine-tunes, but the traffic is concentrated in a handful of corporate releases.
One of three "startups" is just relabelling models. Of 35 startup-tagged entries, the median has 2 K downloads — three orders of magnitude below big-tech medians. Mistral, MiniMax, Moonshot are the visible outliers.
Community fine-tuners specialise in 4–30 B. 59 of 89 community models sit in the 4–30 B range — the sweet spot for consumer GPUs.
2026 is now the platform's biggest release year so far in the dataset. 60 of 300 rows (20 %) carry 2026 publication dates — including the full gemma-4 family (gemma-4-31B-it, gemma-4-26B-A4B-it, gemma-4-E4B-it), DeepSeek's V4 line (DeepSeek-V4-Pro 862 B, DeepSeek-V4-Flash 284 B), and the Qwen3.5 / Qwen3.6 generations. Frontier-model cadence has accelerated since 2025.

Quick start

🤗 Load with `datasets`

from datasets import load_dataset
ds = load_dataset("moncefem/featherless-ai-models", split="train")
print(ds[0])

🐼 Load with `pandas` (no `datasets` install needed)

import pandas as pd
df = pd.read_parquet(
    "hf://datasets/moncefem/featherless-ai-models/featherless_ai_models.parquet"
)
print(df.head())

🦆 Load with DuckDB (in-process SQL)

import duckdb
duckdb.sql("""
  SELECT organisation_name, COUNT(*) AS n, SUM(n_downloads_absolute) AS dl
  FROM 'hf://datasets/moncefem/featherless-ai-models/featherless_ai_models.parquet'
  GROUP BY 1 ORDER BY dl DESC LIMIT 10
""").show()

Charts at a glance

1. Qwen, gemma-4, and Llama dominate the popularity chart

Qwen/Qwen3-0.6B is the most-downloaded model in the catalogue with 18.5 M downloads — bigger by raw count than every other top-15 model combined would be if you removed Qwen and Google. The top 15 is split between two clusters: small-and-cheap workhorses (Qwen3-0.6B, Llama-3.2-1B, gemma-4-E4B, Qwen3-Embedding-0.6B) and mid-size flagships (gemma-4-31B, Qwen3-32B, gpt-oss-20b). Mega-models (200 B+) do not appear in the top 15.

2. The model-size distribution is heavily mid-weight

Bucket	# models	Share
<1 B	15	5 %
1–4 B	38	13 %
4–10 B	110	37 %
10–30 B	67	22 %
30–70 B	28	9 %
70–200 B	23	8 %
200 B+	19	6 %

The 4–10 B bucket alone holds more models than every bucket above 30 B combined. This is the consumer-GPU sweet spot: models that fit on a single RTX 3090/4090 or a Mac Studio. Featherless.ai's catalogue reflects what people actually run, not what gets press coverage.

3. But downloads tell a different story

The 4–10 B bucket leads in absolute downloads (83 M) — but on a per-model basis, <1 B models average 2.0 M downloads, more than 3× any other bucket. Tiny instruction-tuned models (Qwen3-0.6B, Llama-3.2-1B) are the platform's silent traffic kings.

4. Catalogue: a long tail with a Qwen mountain

The 300 models come from 152 distinct organisations — a true long tail. Qwen (41 models), Google (14), Mistral (10), Meta (9) and DeepSeek (7) account for 81 listings = 27 % of the catalogue. The remaining 73 % is spread across 147 organisations averaging 1.5 models each.

5. Aggregate downloads: Qwen alone = half the platform

Qwen's 118 M downloads is 3.1× Google, 4.8× Meta, 9.5× OpenAI in this snapshot. Whoever runs Alibaba's open-weights strategy is winning the attention/distribution game on this platform, irrespective of leaderboard scores. The top 5 orgs (Qwen, Google, Meta, OpenAI, DeepSeek) account for 89 % of all downloads while contributing only 26 % of the models.

6. Org-category x-ray

The catalogue count is community-led (89/215 categorised = 41 %) but the traffic is big-tech-led (180 M of 227 M).

Category	# models	Total downloads	Median downloads/model
Community fine-tuners	89	4.6 M	307
Research labs (Qwen, DeepSeek, Mistral …)	46	130 M	252 500
Big tech (Google, Meta, OpenAI, Microsoft, NVIDIA)	36	92 M	945 500
Startups	35	9.3 M	2 000
Uncategorised	85 + 9	—	—

The 3 000× gap between big-tech and community median downloads is the single biggest finding in this dataset. Hosting a model on Featherless gives community fine-tuners catalogue presence but not traffic: half of community models have fewer than 307 lifetime downloads.

7. Publication timeline — release cadence is accelerating

The platform's catalogue grows roughly an order of magnitude faster every year:

2023: 24 models (the original Llama-2 / Mistral-7B / Qwen-1.5 cohort).
2024: 75 models (the Llama-3.1 / Qwen-2.5 / DeepSeek-V2 era).
2025: 131 models (Qwen-3 wave, gpt-oss release, DeepSeek-V3.x).
2026: 60 models in the first ~5 months alone, including the gemma-4 family, DeepSeek-V4-Pro/Flash, and the Qwen3.5 / Qwen3.6 generations.

Monthly spikes line up with major release events — April 2025 (21 models, Qwen-3 launch + gpt-oss-20b run-up), March 2026 (17 models, gemma-4 family ship), February 2026 (13 models, Qwen3.5).

8. Size vs popularity — almost no relationship

This is the most interesting plot. Each dot is one model, coloured by org category. On log–log axes, the Pearson correlation between parameter count and downloads is r = 0.18 — i.e. size explains only ~3 % of the variance in popularity.

What does explain popularity:

The org name (community vs big-tech accounts for the visible vertical separation).
Instruction-tuned vs base (instruct variants cluster in the upper region).
Being a member of a known model family — Qwen3, Llama-3, gemma-4 all "lift" their entire size-range together.

Bigger does not buy you reach. Branding and ecosystem do.

9. Modality

90 % text, 8 % vision, 1 % embeddings, < 1 % multimodal. Featherless.ai is overwhelmingly a text-LLM platform; vision-language models, embeddings, and multimodal systems are present but rare.

10. Who builds what size of model?

Three patterns jump out:

Community fine-tuners cluster at 4–30 B. 59 / 89 community models sit there — the consumer-GPU sweet spot.
Research labs are the only ones consistently producing 200 B+ models (5 of 19 frontier-class models come from research labs; only 1 from big tech, 0 from community).
Startups skew mid-size (12 + 13 + 3 = 28 of 35 startup models are 4–30 B); few startups go big, almost none go small.

Codebook

Silver — LLM-extracted analytical variables (20)

Column	Type	Description
`id`	string	Deterministic content-hash id
`model_name`	string	Display name (e.g. `Qwen/Qwen3-0.6B`)
`organisation_name`	string	Publishing org slug (e.g. `Qwen`, `google`, `meta-llama`)
`parameter_count_raw`	string	Original "862B" / "8B" / "1.6T" string
`context_length_raw`	string	Original "32K" / "128K" string
`model_type_raw`	string	LLM-stated architecture / family (e.g. `qwen3-32b`, `deepseek4-1.6t`)
`download_count_raw`	string	Original "3.6M" / "222k" string
`published_date`	date	YYYY-MM-DD claimed publication
`n_parameters_billions`	float	Parsed parameter count in billions
`n_context_length_tokens`	integer	Parsed context window in tokens
`n_downloads_absolute`	integer	Parsed download count
`cat_modality`	enum	text · vision · embedding · multimodal
`is_instruction_tuned`	boolean	Variant carries an *-Instruct / -Chat / -it suffix
`is_quantized`	boolean	Page indicates GGUF / GPTQ / AWQ / int8 / fp8
`has_vision_capability`	boolean	Page mentions image / vision input
`cat_org_category`	enum	community · research_lab · big_tech · startup · unknown
`dn_year`, `dn_month`	integer	Decomposed publication date
`is_high_popularity`	boolean	LLM judgement on whether the model is "well-known"
`is_large_model`	boolean	`n_parameters_billions ≥ 70`
`is_long_context`	boolean	`n_context_length_tokens > 32 000`

How this dataset was built

This file was produced by Gemma Miner in a single autonomous agent run.

Harvest — agent fetched the Featherless.ai catalogue pages (10 pages × ~30 models = ~300 listings) and cached the HTML.
Codebook design — an LLM proposed 20 typed variables matching the analytical brief (model size, modality, popularity, org type).
Per-row extraction — for each model card, an LLM read the visible text and emitted a JSON object conforming to the codebook; the system deterministically coerced values (size "862B" → 862.0, downloads "3.6M" → 3_600_000).
Export — parquet + JSONL + this card + ten matplotlib charts.

No fine-tuning. No labelled training data. Reproducible.

🔬 Rebuild this dataset from scratch

pip install gemma-miner
gemma-miner \
  --goal "scrape Featherless.ai's model catalogue and produce a 300-row dataset \
          covering model name, organisation, parameter count, context length, \
          downloads, modality, and organisation type" \
  --url https://featherless.ai \
  --min-rows 300 \
  --required-fields "model_name,organisation_name,parameter_count_raw,download_count_raw"

Limitations & honest caveats

Read these before publishing analysis. The LLM extractor is fast but not infallible.

cat_org_category is missing on 85 / 300 rows (28 %) — these are the long-tail community fine-tuners the LLM couldn't classify. Treat the org-category aggregates as lower bounds.
is_high_popularity is sparse (84 / 300 non-null) — use the numeric n_downloads_absolute instead.
is_quantized is essentially useless (only 3 True values across 109 non-null rows). Featherless.ai doesn't surface quantisation prominently in the catalogue listing.
Download counts are platform-specific. Featherless.ai's download counter is not the same as Hugging Face's — these numbers reflect Featherless usage, not global popularity.
Sample = the first ~10 pages of the catalogue, sorted by Featherless's default ranking (likely popularity). This is not a uniform sample of all hosted models — popular models are massively over-represented compared to the full long tail.
Org tagging is heuristic. Model namespacing (e.g. meta-llama/Llama-3.1-8B-Instruct) was used to attribute the org; re-uploads under a third party's namespace will be misattributed.
No de-duplication across model variants. Qwen3-4B and Qwen3-4B-Instruct-2507 count as two rows.

Citation

@misc{elmouden_featherless_ai_models_2025,
  title  = {Featherless.ai Model Catalog — 300 Models Analysed},
  author = {EL-Mouden, Moncif},
  year   = {2025},
  note   = {Generated by Gemma Miner from https://featherless.ai},
  url    = {https://huggingface.co/datasets/moncefem/featherless-ai-models}
}

@software{elmouden_gemma_miner_2025,
  title  = {Gemma Miner: an autonomous text-to-dataset agent},
  author = {EL-Mouden, Moncif and contributors},
  year   = {2025},
  url    = {https://github.com/moncifem/gemma-miner}
}

Underlying model listings are published by their respective organisations on Featherless.ai; consult those pages for the authoritative model cards.

Author & links

👤 Moncif EL-Mouden — 🤗 huggingface.co/moncefem
🤖 Gemma Miner (the generator) — github.com/moncifem/gemma-miner
🪶 Source — featherless.ai

License

Apache License 2.0.

Please attribute:

Featherless.ai as the source of the underlying catalogue, and
Gemma Miner (https://github.com/moncifem/gemma-miner) as the dataset generator.

Downloads last month: 123