Evaluation · Registry

Model registry

Facts about current AI models, limited to what their developers state on official pages — each value linked to the page it was confirmed on.

Models: 14
Organizations: 8
Last verified: 28 Sept 2026

Compare models Benchmarks JSON

OrganizationAll OpenAI0 Anthropic0 Google DeepMind1 Meta2 Mistral AI1 DeepSeek1 Alibaba Cloud (Qwen)1 xAI0

WeightsAll Open weights6 API-served6

InputAny Text8 Image6 Audio0 Video1 PDF0

Showing 6 of 14. Clear filters

Models matching Open weights, Image input (6 of 14), newest first

Model	Organization	Released	Context	Input modalities	Weights
DeepSeek-V4.1-Flash`deepseek-flash`	DeepSeek	10 Sept 2026Announced	1,000,000 tokens	Text, Image	Open weights
Qwen3.8-27B`qwen3.8-27b`	Alibaba Cloud (Qwen)	14 Aug 2026Released	262,144 tokens	Text, Image, Video	Open weights
Muse Glimmer	Meta	10 Aug 2026Announced	131,072 tokens	Text, Image	Open weights
Mistral Medium 3.5`mistral-medium-3-5`	Mistral AI	28 Apr 2026Released	262,144 tokens	Text, Image	Open weights
Gemma 4 31B`gemma-4-31b-it`	Google DeepMind	31 Mar 2026Released	262,144 tokens	Text, Image	Open weights
Llama 4 Maverick	Meta	5 Apr 2025Released	1,000,000 tokens	Text, Image	Open weights

Only fields confirmed on an official page from the model’s developer are shown; each model page links the exact source for every value. “Not verified” means we could not confirm it — not zero, and not unknown to the vendor.

Context is the context window in tokens, abbreviated (1M = 1,000,000). Some vendors state an input-token limit instead; each model page gives the exact figure and the vendor’s own wording. “API-served” means the entry describes a hosted model reached through the vendor’s API. The full registry is available as JSON.