Every model that matters, current and cutting edge.

Anthropic's new top tier above Opus, with always-on adaptive thinking and 1M context.

Top-tier reasoning model in the Claude family. Highest intelligence, slower and more expensive than Sonnet.

Workhorse tier — strong intelligence at lower latency/cost than Opus. The default for most production use.

Fast, low-cost tier. Best for high-volume, latency-sensitive tasks.

Multimodal flagship of the 4-series.

Successor to 4o focused on coding and instruction.

Frontier reasoning flagship.

Reasoning-first inference-time-compute models (o1, o3, o4).

Cost-efficient multimodal tier.

OpenAI's open-weight models (Apache-2.0).

Workhorse multimodal Gemini tier.

Top reasoning Gemini tier.

Fast, low-cost multimodal tier.

On-device Gemini.

Open-weights Google models.

Predecessor to Gemini (legacy).

Frontier model family from xAI.

Open-weights releases of older Grok generations.

Llama text/instruction families (Llama 2 -> 4).

Code-specialized Llama 2 derivative.

Safety classifier built on Llama.

Original open Mistral.

MoE open-weight family.

Flagship proprietary tier.

Mid-tier cost-efficient.

Code-specialized.

Vision-language open weights.

Edge/on-device tier.

Reasoning-focused tier.

Mistral's speech generation family (TTS).

Agentic coding family (merged into Small 4 in March 2026).

General dense LLMs.

Code models (dense and MoE).

MoE flagship family.

Reasoning models trained with RL.

Vision-language family.

General Qwen dense LLMs.

Qwen MoE flagships (Qwen3 family).

Code-specialized Qwen.

Vision-language Qwen.

Audio-language Qwen.

Math-specialized Qwen.

Any-to-any multimodal Qwen.

Reasoning Qwen.

Open and proprietary GLM series.

Code-specialized GLM.

Vision-language GLM.

Long-context chat models from Moonshot.

Open-weights Kimi releases.

Dense Yi models.

Vision-language Yi.

Code Yi.

Closed flagship (Lightning).

Open and proprietary Baichuan models.

MoE / linear-attention frontier; some open weights.

Closed flagship multimodal models.

Closed Hunyuan models.

Open MoE Hunyuan.

Open video generation model.

Open-source 3D world models.

ByteDance Seed/Doubao 2.0 agentic series.

ByteDance Doubao series (Volc Engine).

Closed Command series.

Open-weights Command for RAG/tools.

Multilingual open-weights research models.

Multilingual embeddings.

Small, capable open-weights from Microsoft Research.

Hybrid Mamba-Transformer.

Multimodal Reka models (Edge / Flash / Core).

Personal AI; closed.

Open-weights MoE from Databricks.

Open-weights enterprise models from IBM Research.

Open enterprise MoE.

NVIDIA's third-gen open Nemotron family.

Open NVIDIA-trained models (often Llama-based).

Diffusion-Transformer architecture; native 4K, dedicated text-rendering module.

Open image generation.

Open video generation.

Open / API image generation from former Stability core.

Open-weights Falcon series.

Search-augmented chat models.

Amazon Bedrock foundation models.

Legacy Amazon Titan models.

On-device + server foundation models powering Apple Intelligence.

Open small models from Apple research.

Groq operates inference for upstream open models; this catalog tracks Groq-original fine-tunes only.

Models built via evolutionary model-merge.

Community fine-tunes on top of open base models.