name stringlengths 5 29 | slug stringlengths 5 29 | convly_url stringlengths 30 54 | developer stringlengths 4 11 | model_type stringlengths 3 25 | modality stringclasses 4
values | parameters stringlengths 2 30 | context_window stringclasses 8
values | max_output stringclasses 6
values | license stringclasses 9
values | open_weights stringclasses 2
values | release_date stringclasses 6
values | input_price float64 0.02 10 ⌀ | output_price float64 0.03 50 ⌀ | api_providers stringlengths 13 39 | vram_q4 stringlengths 5 14 ⌀ | min_gpu stringlengths 11 40 ⌀ | official_url stringlengths 22 37 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Claude Fable 5 | claude-fable-5 | https://convly.ai/model/claude-fable-5/ | Anthropic | LLM (frontier reasoning) | Text, Vision → Text | Undisclosed | 1M | 128K | Proprietary | no | 2026 | 10 | 50 | Anthropic, AWS | null | null | https://www.anthropic.com/claude |
Claude Haiku 4.5 | claude-haiku-4-5 | https://convly.ai/model/claude-haiku-4-5/ | Anthropic | LLM | Text, Vision → Text | Undisclosed | 200K | 64K | Proprietary | no | 2025 | 1 | 5 | Anthropic, AWS, Vertex AI, Azure | null | null | https://www.anthropic.com/claude |
Claude Opus 4.8 | claude-opus-4-8 | https://convly.ai/model/claude-opus-4-8/ | Anthropic | LLM (reasoning) | Text, Vision → Text | Undisclosed | 1M | 128K | Proprietary | no | 2026 | 5 | 25 | Anthropic, AWS, Vertex AI, Azure | null | null | https://www.anthropic.com/claude |
Claude Sonnet 4.6 | claude-sonnet-4-6 | https://convly.ai/model/claude-sonnet-4-6/ | Anthropic | LLM | Text, Vision → Text | Undisclosed | 1M | 64K | Proprietary | no | 2025 | 3 | 15 | Anthropic, AWS, Vertex AI, Azure | null | null | https://www.anthropic.com/claude |
DeepSeek R1 | deepseek-r1 | https://convly.ai/model/deepseek-r1/ | DeepSeek | LLM (MoE, reasoning) | Text → Text | 671B total / 37B active (MoE) | 128K | null | MIT (open) | yes | 2025 | 0.5 | 2.15 | DeepSeek, DeepInfra, OpenRouter | ~400 GB | Multi-GPU server | https://huggingface.co/deepseek-ai |
DeepSeek R1 Distill Llama 70B | deepseek-r1-distill-llama-70b | https://convly.ai/model/deepseek-r1-distill-llama-70b/ | DeepSeek | LLM (dense, reasoning) | Text → Text | 70B | 128K | null | MIT (open) | yes | 2025 | 0.8 | 0.8 | DeepInfra, OpenRouter, Ollama | ~40 GB | 2× RTX 4090 / 1× 48GB | https://huggingface.co/deepseek-ai |
DeepSeek V4-Flash | deepseek-v4-flash | https://convly.ai/model/deepseek-v4-flash/ | DeepSeek | LLM (MoE) | Text → Text | 284B total / ~13B active (MoE) | 1M | 384K | MIT (open) | yes | 2026-04 | 0.14 | 0.28 | DeepSeek, OpenRouter | ~140 GB | 2× H100 80GB (4-bit) | https://huggingface.co/deepseek-ai |
DeepSeek V4-Pro | deepseek-v4-pro | https://convly.ai/model/deepseek-v4-pro/ | DeepSeek | LLM (MoE) | Text → Text | 1.6T total / ~49B active (MoE) | 1M | 384K | MIT (open) | yes | 2026-04 | 0.435 | 0.87 | DeepSeek, OpenRouter | ~800 GB | Multi-GPU server (e.g. 8× H100 80GB) | https://huggingface.co/deepseek-ai |
Gemini 3.1 Pro | gemini-3-1-pro | https://convly.ai/model/gemini-3-1-pro/ | Google | LLM (multimodal) | Text, Image, Audio, Video → Text | Undisclosed | 1.05M | 65K | Proprietary | no | 2026 | 2 | 12 | Google AI Studio, Vertex AI | null | null | https://ai.google.dev/ |
Gemini 3.5 Flash | gemini-3-5-flash | https://convly.ai/model/gemini-3-5-flash/ | Google | LLM (multimodal) | Text, Image, Audio, Video → Text | Undisclosed | 1M | 65K | Proprietary | no | 2026 | 1.5 | 9 | Google AI Studio, Vertex AI | null | null | https://ai.google.dev/ |
Gemma 3 12B | gemma-3-12b | https://convly.ai/model/gemma-3-12b/ | Google | LLM (multimodal) | Text, Image → Text | 12B | 128K | null | Gemma (open) | yes | 2025 | 0.05 | 0.15 | Google AI Studio, OpenRouter, Ollama | ~8 GB | RTX 4070 12GB | https://huggingface.co/google |
Gemma 3 27B | gemma-3-27b | https://convly.ai/model/gemma-3-27b/ | Google | LLM (multimodal) | Text, Image → Text | 27B | 128K | null | Gemma (open) | yes | 2025 | 0.08 | 0.16 | Google AI Studio, OpenRouter, Ollama | ~16 GB | RTX 4080 16GB / RTX 4090 | https://huggingface.co/google |
Gemma 3 4B | gemma-3-4b | https://convly.ai/model/gemma-3-4b/ | Google | LLM (multimodal) | Text, Image → Text | 4B | 128K | null | Gemma (open) | yes | 2025 | 0.05 | 0.1 | Google AI Studio, Ollama | ~3 GB | Any 6GB+ GPU | https://huggingface.co/google |
GLM 5.2 | glm-5-2 | https://convly.ai/model/glm-5-2/ | Zhipu AI | LLM (coding/agentic, MoE) | Text → Text | 744B total / ~40B active (MoE) | 1M | 131K | MIT (open) | yes | 2026-06 | 1.4 | 4.4 | Zhipu (Z.ai), OpenRouter | ~370 GB | Multi-GPU server (e.g. 5× H100 80GB) | https://github.com/zai-org/GLM-5 |
GPT-5.5 | gpt-5-5 | https://convly.ai/model/gpt-5-5/ | OpenAI | LLM (reasoning) | Text, Vision → Text | Undisclosed | 1.05M | 128K | Proprietary | no | 2026 | 5 | 30 | OpenAI, Azure | null | null | https://openai.com/api/ |
Kimi K2.7 Code | kimi-k2-7-code | https://convly.ai/model/kimi-k2-7-code/ | Moonshot AI | LLM (coding, MoE) | Text → Text | 1T total / 32B active (MoE) | 256K | — | Modified MIT (open) | yes | 2026-06 | 0.6 | 2.5 | Moonshot, OpenRouter | ~500 GB | Multi-GPU server | https://huggingface.co/moonshotai |
Llama 3.1 8B | llama-3-1-8b | https://convly.ai/model/llama-3-1-8b/ | Meta | LLM (dense) | Text → Text | 8B | 128K | null | Llama 3.1 Community (open) | yes | 2024 | 0.02 | 0.03 | Together, DeepInfra, OpenRouter, Ollama | ~5 GB | Any 8GB GPU | https://www.llama.com/ |
Llama 3.3 70B | llama-3-3-70b | https://convly.ai/model/llama-3-3-70b/ | Meta | LLM (dense) | Text → Text | 70B | 128K | null | Llama 3.3 Community (open) | yes | 2024 | 0.1 | 0.32 | Together, DeepInfra, OpenRouter, Ollama | ~40 GB | 2× RTX 4090 / 1× 48GB | https://www.llama.com/ |
Llama 4 Maverick | llama-4-maverick | https://convly.ai/model/llama-4-maverick/ | Meta | Multimodal (MoE) | Text, Image → Text | 400B total / 17B active (MoE) | 1M | null | Llama 4 Community (EU-restricted) | yes | 2025 | 0.15 | 0.6 | Meta, Together, OpenRouter | ~240 GB | Multi-GPU server | https://www.llama.com/models/llama-4/ |
Llama 4 Scout | llama-4-scout | https://convly.ai/model/llama-4-scout/ | Meta | Multimodal (MoE) | Text, Image → Text | 109B total / 17B active (MoE) | 10M | null | Llama 4 Community (EU-restricted) | yes | 2025 | 0.1 | 0.3 | Meta, Together, OpenRouter | ~65 GB | H100 80GB / Mac 128GB | https://www.llama.com/models/llama-4/ |
Mistral 7B | mistral-7b | https://convly.ai/model/mistral-7b/ | Mistral AI | LLM (dense) | Text → Text | 7B | 32K | null | Apache 2.0 (open) | yes | 2023 | 0.02 | 0.03 | Mistral, DeepInfra, OpenRouter, Ollama | ~4.5 GB | Any 6GB GPU | https://huggingface.co/mistralai |
Mistral Large 3 | mistral-large-3 | https://convly.ai/model/mistral-large-3/ | Mistral AI | LLM (MoE) | Text → Text | 675B total / 41B active (MoE) | 256K | null | Apache 2.0 (open) | yes | 2025 | 2 | 6 | Mistral, OpenRouter | ~400 GB | Multi-GPU server | https://huggingface.co/mistralai |
Mistral NeMo 12B | mistral-nemo-12b | https://convly.ai/model/mistral-nemo-12b/ | Mistral AI | LLM (dense) | Text → Text | 12B | 128K | null | Apache 2.0 (open) | yes | 2024 | 0.02 | 0.04 | Mistral, OpenRouter, Ollama | ~7.5 GB | RTX 4070 12GB / RTX 3060 | https://huggingface.co/mistralai |
NVIDIA Nemotron 3 Nano Omni | nvidia-nemotron-3-nano-omni | https://convly.ai/model/nvidia-nemotron-3-nano-omni/ | NVIDIA | Multimodal (omni) | Text, Image, Audio, Video → Text | 30B total / ~3B active (MoE) | 256K | — | NVIDIA Open Model Agreement | yes | 2026 | null | null | Hugging Face, OpenRouter, NVIDIA NIM | ~21 GB (NVFP4) | RTX 5090 32GB (NVFP4) / H100 80GB (BF16) | https://huggingface.co/nvidia |
Phi-4 | phi-4 | https://convly.ai/model/phi-4/ | Microsoft | LLM (dense) | Text → Text | 14B | 16K | null | MIT (open) | yes | 2025 | 0.07 | 0.14 | Azure, OpenRouter, Ollama | ~9 GB | RTX 4070 12GB / RTX 3060 12GB | https://huggingface.co/microsoft |
Qwen3 14B | qwen3-14b | https://convly.ai/model/qwen3-14b/ | Alibaba | LLM (dense) | Text → Text | 14B | 128K | null | Apache 2.0 (open) | yes | 2025 | 0.12 | 0.24 | Alibaba, OpenRouter, Ollama | ~9 GB | RTX 4070 12GB (Q4) | https://huggingface.co/Qwen |
Qwen3 235B-A22B | qwen3-235b-a22b | https://convly.ai/model/qwen3-235b-a22b/ | Alibaba | LLM (MoE) | Text → Text | 235B total / 22B active (MoE) | 128K | null | Apache 2.0 (open) | yes | 2025 | 0.45 | 1.8 | Alibaba, OpenRouter | ~140 GB | Multi-GPU or Mac 192GB | https://huggingface.co/Qwen |
Qwen3 30B-A3B | qwen3-30b-a3b | https://convly.ai/model/qwen3-30b-a3b/ | Alibaba | LLM (MoE) | Text → Text | 30B total / 3B active (MoE) | 128K | null | Apache 2.0 (open) | yes | 2025 | 0.12 | 0.5 | Alibaba, OpenRouter, Ollama | ~18 GB | RTX 4090 24GB (Q4) — fast, 3B active | https://huggingface.co/Qwen |
Qwen3 32B | qwen3-32b | https://convly.ai/model/qwen3-32b/ | Alibaba | LLM (dense) | Text → Text | 32B | 128K | null | Apache 2.0 (open) | yes | 2025 | 0.08 | 0.28 | Alibaba, OpenRouter, Ollama | ~20 GB | RTX 4090 24GB (Q4) | https://huggingface.co/Qwen |
Qwen3 8B | qwen3-8b | https://convly.ai/model/qwen3-8b/ | Alibaba | LLM (dense) | Text → Text | 8B | 128K | null | Apache 2.0 (open) | yes | 2025 | 0.04 | 0.14 | Alibaba, OpenRouter, Ollama | ~5 GB | RTX 3060 8GB / any 8GB GPU | https://huggingface.co/Qwen |
Convly AI Models Database
A continuously updated, hand-verified dataset of 30+ AI language models — specs, licenses, API pricing (USD per 1M tokens), and local-hardware (VRAM) requirements.
Maintained by Convly.ai · Live interactive version: https://convly.ai/models/
Fields
name, slug, convly_url, developer, model_type, modality, parameters, context_window, max_output, license, open_weights, release_date, input_price (USD/1M tokens), output_price (USD/1M tokens), api_providers, vram_q4 (GB, 4-bit), min_gpu, official_url
Coverage
Frontier + open-weight models as of July 2026: Claude (Opus 4.8, Fable 5, Sonnet 4.6, Haiku 4.5), GPT-5.5, Gemini 3.1 Pro / 3.5 Flash, DeepSeek V4-Pro / V4-Flash / R1, Qwen3 family, Llama 3.x / 4, Gemma 3, Mistral, Phi-4, GLM 5.2, Kimi K2.7 Code, NVIDIA Nemotron 3, and more.
Why
Provider pricing pages are inconsistent and marketing-heavy. This dataset normalizes verified specs, like-for-like per-million-token pricing, and practical 4-bit VRAM requirements so you can tell instantly whether a model runs on your hardware and what it costs to call. It powers the LLM VRAM Calculator, the AI API Cost Calculator, and the AI Price-Performance Index.
Usage
from datasets import load_dataset
ds = load_dataset("sakd99/ai-models-database")
print(ds["train"][0])
Or load the raw files directly:
import pandas as pd
df = pd.read_csv("convly-ai-models.csv")
License
CC BY 4.0 — free to use, share and adapt with attribution to Convly.ai (a link to https://convly.ai/models/ is perfect).
Citation
Convly.ai (2026). Convly AI Models Database [Data set]. https://convly.ai/models/
- Downloads last month
- 11