Models
20 entries
| Entry | Summary | As of |
|---|---|---|
| AlphaFoldprotein-structure prediction system | AlphaFold's landmark report describes highly accurate protein-structure prediction with a neural-network system. Predicted structure confidence and experimental validation remain distinct forms of evidence. | 2026-07-13 |
| BERTencoder language model | BERT pretrains bidirectional Transformer encoders using masked-token and sentence-level objectives. Its architecture is oriented toward learned representations for downstream tasks rather than autoregressive generation. | 2026-07-13 |
| CLIPvision-language model | CLIP learns visual representations from natural-language supervision across image-text pairs. Its zero-shot classification method compares image and text representations rather than training a new classifier per benchmark. | 2026-07-13 |
| Claude model familylanguage-model family | Claude is a versioned family with models optimized for different capability, latency, and cost profiles. Anthropic publishes system cards alongside model documentation to record evaluation and safety evidence. | 2026-07-13 |
| Command model familyenterprise language-model family | Cohere's Command family targets enterprise generation, retrieval, and tool-use workloads. The exact variant matters because context, tool support, deployment, and price can differ. | 2026-10-26 |
| DALL-Etext-to-image model family | The original DALL-E research framed text-to-image generation as modeling image tokens conditioned on text. Later products carrying the name are versioned systems and should not be assumed identical to the original paper. | 2026-07-13 |
| DeepSeek-V3mixture-of-experts language model | DeepSeek-V3 is described in its technical report as a mixture-of-experts language model. The report documents architecture and training choices; deployment terms must be checked separately from research access. | 2026-07-13 |
| Diffusion modelsgenerative model family | Denoising diffusion probabilistic models learn to reverse a gradual noising process. Generation begins from noise and iteratively applies learned denoising transitions. | 2026-07-13 |
| GPT model familylanguage-model family | GPT is a changing OpenAI model family rather than one fixed system. The provider's model catalog is the current source for supported modalities, endpoints, and lifecycle status. | 2026-07-13 |
| Gemini model familymultimodal model family | Gemini is Google's versioned family of multimodal developer models. Model variants differ in supported inputs, context, endpoints, and lifecycle, so comparisons need an as-of date. | 2026-07-13 |
| Grok model familylanguage-model family | Grok is a versioned family exposed by xAI through documented API identifiers. Aliases are convenient but pinned IDs and evaluation dates are stronger evidence for repeatable comparisons. | 2026-07-13 |
| Llama model familyopen-weight model family | Llama is Meta's downloadable model family distributed under release-specific community licenses. Open weights describe access to parameters; they do not by themselves establish an OSI-approved open-source license. | 2026-07-13 |
| Mistral model familylanguage-model family | Mistral documents general, reasoning, coding, and multimodal model variants. A family-level label is insufficient for reproducibility; the exact model ID and date belong in an evaluation record. | 2026-07-13 |
| Mixture-of-experts modelssparse model architecture | A sparsely gated mixture-of-experts layer activates only selected expert subnetworks for each input. Conditional computation can increase parameter capacity without using every parameter on every token. | 2026-07-13 |
| Qwen model familymultimodal model family | Qwen encompasses language, vision-language, coding, and other task-focused model releases. Model cards and repository licenses should be inspected for each checkpoint before commercial reuse. | 2026-07-13 |
| Soravideo-generation model | OpenAI's Sora technical report presents video generation as a route toward general-purpose world simulation. The report is a provider-authored technical account, not an independent benchmark of physical understanding. | 2026-07-13 |
| Stable Diffusionlatent-diffusion model family | Stable Diffusion refers to a family of image-generation models and related hosted products. Checkpoint licenses and hosted-service terms can differ, so the family name does not establish reuse rights. | 2026-07-13 |
| T5text-to-text model | T5 casts diverse natural-language tasks into a unified text-to-text format. The research systematically studies transfer learning choices using the same interface across tasks. | 2026-07-13 |
| Vision Transformervision model | Vision Transformer applies a Transformer encoder to sequences of image patches. The paper shows that attention-based image models can compete strongly when pretrained at sufficient scale. | 2026-07-13 |
| Whisperspeech model | Whisper is an encoder-decoder Transformer trained on large-scale weakly supervised speech data. The research reports multilingual transcription and translation experiments, not a guarantee for every accent or acoustic setting. | 2026-07-13 |