Every model, measured.

Start with a research area, drill into a vendor, or page through the full index. Only models with at least one benchmark score appear — a model without a recorded score can’t be ranked.

Vendor:Areas overview speakleash · 253 OpenAI · 85 Google · 71 Qwen · 52 Alibaba · 47 Anthropic · 44 Microsoft · 35 Meta · 30 Mistral · 30 DeepSeek · 28 google · 19 meta-llama · 19 mistralai · 19 Meta AI · 15 CYFRAGOVPL · 14 Zhipu AI · 13 NVIDIA · 10 SpeakLeash · 10 internlm · 10 xAI · 10 ByteDance · 9 Baidu · 8 PLLuM · 8 ibm-granite · 8 microsoft · 8 Amazon · 7 Google DeepMind · 7 MiniMax · 7 Mistral AI · 7 Remek · 7 Shanghai AI Lab · 7 allenai · 7 utter-project · 7 CohereForAI · 6 Microsoft Research · 6 Salesforce · 6 01-ai · 5 Alibaba Cloud · 5 Cohere · 5 Moonshot AI · 5 NousResearch · 5 THUML · 5 deepseek-ai · 5 DeepMind · 4 Facebook AI · 4 IBM · 4 Meituan · 4 Stanford · 4 THUDM · 4 UC San Diego · 4 VikParuchuri · 4 gguf-iq · 4 nvidia · 4 openchat · 4 tiiuae · 4 Allen AI · 3 BAAI · 3 Du et al. · 3 ForgeCode · 3 Fudan University · 3 IDEA Research · 3 Liao et al. · 3 Moonshot.AI · 3 Nam Tuan Ly / NII · 3 OPI-PG · 3 OpenDataLab · 3 ViCoS Lab Ljubljana · 3 Xiaomi · 3 Zhao et al. · 3 gguf · 3 gguf11bv30 · 3 gguf7bv30 · 3 upstage · 3+ 247 smaller vendors (291 models)

§ 01 · Microsoft models

35 models from Microsoft · page 1 of 1.

#	Model	Vendor	Parameters	Architecture	SOTA	Benchmarks	Results
001	DeBERTa-v3-large	Microsoft	304M	DeBERTa-v3-large	3	5	6
002	Phi-4	Microsoft	14B	transformer	2	3	17
003	RAD-DINO	Microsoft	—	Self-supervised ViT	1	2	2
004	VALL-E 2	Microsoft	Unknown	Neural codec language model (EnCodec tokens)	1	2	2
005	NaturalSpeech 3	Microsoft	~500M	Factorized codec + non-AR diffusion	1	1	1
006	Swin Transformer V2 Large	Microsoft	197M	Hierarchical Vision Transformer	1	1	1
007	WavLM Large (SV)	Microsoft	316M	WavLM Large + ECAPA-TDNN head	1	1	1
008	Phi-4 Multimodal Instruct	Microsoft	6B	Phi-4 multimodal	—	8	9
009	WizardLM-2-8x22b	Microsoft	—	—	—	1	7
010	E5-Mistral-7B-instruct	Microsoft	7B	Mistral-7B (LLM-based embedding)	—	3	4
011	ResNet-152	Microsoft	60M	CNN	—	2	3
012	ResNet-50	Microsoft	25M	CNN	—	3	3
013	UniXcoder	Microsoft	Unknown	Transformer encoder-decoder	—	3	3
014	CodeBERT	Microsoft	Unknown	BERT	—	2	2
015	Florence-2-Large	Microsoft	—	—	—	1	2
016	KOSMOS-2.5	Microsoft	—	—	—	1	2
017	LightGBM	Microsoft	—	Gradient Boosted Trees (leaf-wise)	—	2	2
018	Azure Document Intelligence	Microsoft	Unknown	Managed layout + OCR extraction service	—	1	1
019	Azure OCR	Microsoft	—	Cloud OCR Service	—	1	1
020	BEiT-3 (ViT-L)	Microsoft	Unknown	Multiway Transformer (ViT-L/14)	—	1	1
021	BioViL	Microsoft	—	Vision-Language Transformer	—	1	1
022	CodeBERT	Microsoft	—	BERT pretrained on code + NL	—	1	1
023	DeBERTa (ensemble)	Microsoft	—	—	—	1	1
024	DiT-Base	Microsoft	—	Vision Transformer (self-supervised)	—	1	1
025	DiT-Large	Microsoft	Unknown	Document Image Transformer Large	—	1	1
026	GraphCodeBERT	Microsoft	125M	transformer	—	1	1
027	LayoutLMv3	Microsoft	Unknown	Multimodal Transformer (text + layout + image)	—	1	1
028	Pengi	Microsoft	~300M	CLAP audio encoder + GPT-2 decoder	—	1	1
029	Phi-4 14B	Microsoft	14B	—	—	1	1
030	Swin Transformer Large	Microsoft	197M	Hierarchical Vision Transformer	—	1	1
031	Swin-L + UperNet	Microsoft	Unknown	Swin Transformer Large backbone + UperNet head	—	1	1
032	UFO (GPT-4V)	Microsoft	Unknown	UI-Focused agent with dual-agent architecture on GPT-4V	—	1	1
033	VALL-E	Microsoft	~400M	Neural codec LM (EnCodec tokens)	—	1	1
034	mDeBERTa-v3-base	Microsoft	86M	DeBERTa-v3 (multilingual)	—	1	1
035	swin_large.ms_in22k_ft_in1k	Microsoft	—	Swin-L, IN22K pre-train, IN1K fine-tune	—	1	1