Open-Weight Models

Pyyan / Open-Weight Models

24 entries

Downloadable weights you can self-host. · ranked by benchmark performance and adoption · verified 13 Aug 2026

#	Name	Context	Input ($/Mtok)	Output ($/Mtok)	Licence	Max output	Status
1	DeepSeek V4-Pro DeepSeek	—	—	—	MIT	—	current
2	Kimi K3 Moonshot AI	—	—	—	Open weights	—	current
3	GLM-5.2 Zhipu AI	—	—	—	Open weights	—	current
4	Qwen3-235B-A22B Alibaba	—	—	—	Apache-2.0	—	current
5	Llama 5 Meta	—	—	—	Llama licence	—	current
6	Hunyuan 3.0 Tencent	256K	—	—	Apache-2.0	—	current
7	Qwen3-Coder-480B Alibaba	—	—	—	Apache-2.0	—	current
8	MiniMax M3 MiniMax	—	—	—	Open weights	—	current
9	gpt-oss-120b OpenAI	—	—	—	Open weights	—	current
10	Nemotron 3 Ultra NVIDIA	1M	—	—	NVIDIA Open Model	—	current
11	Kimi K2.6 Moonshot AI	—	—	—	Open weights	—	current
12	GLM-5.1 Zhipu AI	—	—	—	Open weights	—	current
13	DeepSeek V3.2 DeepSeek	—	—	—	MIT	—	current
14	Llama 4 Maverick Meta	—	—	—	Llama licence	—	current
15	Mistral Large 3 Mistral	—	—	—	Apache-2.0	—	current
16	Phi-4 Microsoft	—	—	—	MIT	—	current
17	Gemma 3 27B Google	—	—	—	Gemma licence	—	current
18	TFalcon TII	—	—	—	Open weights	—	current
19	ERNIE 4.5 Baidu	—	—	—	Open weights	—	current
20	gpt-oss-20b OpenAI	—	—	—	Open weights	—	current
21	Phi-4-mini Microsoft	—	—	—	MIT	—	current
22	Gemma 3 4B Google	—	—	—	Gemma licence	—	current
23	Step 3.5 Flash StepFun	—	—	—	Open weights	—	current
24	SSarvam Sarvam AI	—	—	—	Open weights	—	current

DeepSeek

80.6%

SWE-Bench Verified

Moonshot AI

2.8T

parameters