Skip to content

Evidence-backed directory

AI Models Directory

Verified model identities and model-card facts, with unknown fields left unknown.

16 public recordsPrimary-source publication gateCanonical URLs only

model

Sarvam-30B

Sarvam AI

A Mixture-of-Experts text-generation model with a strong India-language focus. Model-card-reported benchmark results are not promoted as independent rankings.

verifiedmodelindiclocal

model

Sarvam-105B

Sarvam AI

Large Sarvam model with a first-party model card and downloadable artifacts; publisher benchmarks remain labeled as publisher-reported.

verifiedmodelindiclocal

Base model focused on Hindi, English and Hinglish; the source explicitly notes it is a base model rather than a ready-made chat assistant.

verifiedmodelindichindi

model

Airavata

AI4Bharat

Hindi instruction-tuned model built from OpenHathi and the IndicInstruct dataset.

verifiedmodelindichindi

model

Krutrim-2-Instruct

OLA Krutrim

Instruction-tuned 12B model built on Mistral-NeMo architecture with an India/Indic focus.

verifiedmodelindiclocal

model

Bark

Suno

Open-source model/tooling project with a public, non-archived upstream repository (suno-ai/bark). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

Whisper

OpenAI

Open-source model/tooling project with a public, non-archived upstream repository (openai/whisper). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

Faster-Whisper

SYSTRAN (open source)

Open-source model/tooling project with a public, non-archived upstream repository (SYSTRAN/faster-whisper). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

WhisperX

WhisperX contributors

Open-source model/tooling project with a public, non-archived upstream repository (m-bain/whisperX). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

Silero Models

Silero AI

Open-source model/tooling project with a public, non-archived upstream repository (snakers4/silero-models). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

MiniCPM 3B

ModelBest

Open-source model project with a public, non-archived upstream repository (OpenBMB/MiniCPM). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

Intern 7B

Alibaba

Open-source model project with a public, non-archived upstream repository (InternLM/InternLM). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

DeepSeek-V2

DeepSeek

Open-source model project with a public, non-archived upstream repository (deepseek-ai/DeepSeek-V2). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

ChatGLM

Zhipu AI / Tsinghua University

Open-source model project with a public, non-archived upstream repository (zai-org/ChatGLM3). Capability and benchmark claims are not inherited from repository identity and remain unknown unless separately evidenced.

verifiedmodel

model

StarCoder2

BigCode (Hugging Face, ServiceNow, NVIDIA)

Open-weight model with a public, non-disabled model card on Hugging Face (bigcode/starcoder2-15b). Capability and benchmark claims are not inherited from model-card identity and remain unknown unless separately evidenced.

verifiedmodel

model

Zephyr

Hugging Face

Open-weight model with a public, non-disabled model card on Hugging Face (HuggingFaceH4/zephyr-7b-beta). Capability and benchmark claims are not inherited from model-card identity and remain unknown unless separately evidenced.

verifiedmodel

How model verification differs from agent verification

A model card's identity is verified against its official publisher page — usually a Hugging Face model card or the developer's own documentation. Benchmark numbers reported on that page are labeled as publisher-reported, not treated as independent rankings; this site does not re-run benchmarks or convert publisher claims into comparative scores without separate reproducible evidence.

Imported catalog coverage

These 72 models are listed for directory coverage while field-level evidence is verified. Their identity is not yet independently confirmed, so no capability or pricing claim is attached.