AI Fact Checker
Directory
Companies
Who builds today's AI — models, harnesses, and the tools that run them
All
Language
Agent Harness
Local Runtime
Image
Video
Audio & Music
Scientific AI
Computer Vision
AI Product
23 companies
85 models
204 sources
OpenAI
17 models
OpenAI is an AI research laboratory that develops frontier AI models including the GPT series. Known for
ChatGPT
, they produce both general-purpose and specialized reasoning models. Their current flagship family is
GPT-5.6
, released in three tiers: Sol (frontier), Terra (balanced), and Luna (fast and affordable).
GPT-5.6 Sol
GPT-5.4
GPT-5.4 mini
+14
Anthropic
11 models
Anthropic is an AI safety company developing the
Claude
family of models, with a strong emphasis on safety research and Constitutional AI. Their
Claude
5 series — Sonnet 5,
Fable 5
,
Mythos 5
, and Opus 5 — shipped in rapid succession in mid-2026, with Opus 5 as the everyday default and
Fable 5
at the research frontier.
Claude Fable 5
Claude Opus 5
Claude Sonnet 5
+8
Google DeepMind
10 models
Google DeepMind develops the
Gemini
family of models, pushing boundaries in
multimodal
AI, long-context understanding, and reasoning. Their Flash tier ships frequently (
Gemini 3.6 Flash
arrived July 2026), while
Gemini
3.5 Pro remains delayed for an architectural rebuild and
Gemini
4 is in
pre-training
.
Gemini 3.6 Flash
Gemini 2.5 Pro
Gemini 2.5 Flash
+7
Meta AI
8 models
Meta AI (now under Meta Superintelligence Labs) develops both the open-weight
Llama
family and the newer closed Muse family. In July 2026 Meta returned to the frontier with
Muse Spark 1.1
and launched its first-ever paid developer
API
— a notable shift from its open-
weights
-only strategy.
Muse Spark 1.1
Llama 4 Scout
Llama 4 Maverick
+5
DeepSeek
4 models
DeepSeek
is a Chinese AI company that has disrupted the industry with ultra-efficient, open-weight models. Their April 2026 V4 family brought a 1M-
token
context window
as standard, with the 1.6T-
parameter
V4 Pro briefly the largest open-weight model ever released — at prices far below any Western frontier model.
DeepSeek V4
DeepSeek V3
DeepSeek R1
+1
Moonshot AI
3 models
Moonshot AI is a Beijing-based lab best known for its
Kimi
assistant and the open-weight K-series models. In July 2026 it released
Kimi K3
— at 2.8 trillion
parameters
, the largest open-weight model ever published — which debuted #1 on the
Frontend Code Arena
above
Claude Fable 5
.
Kimi K3
Kimi K2.6
Kimi K2
xAI
5 models
xAI, founded by Elon Musk, develops the
Grok
family of models. In July 2026 xAI was dissolved into the unified SpaceXAI brand, and
Grok 4.5
launched days later as its first flagship — trained with
Cursor agent
-interaction data and pitched on efficiency rather than raw scale.
Grok 4.5
Grok 4.20
Grok 4.1 Fast
+2
Mistral AI
5 models
Mistral
AI is a French AI company building efficient, open-weight models. Their April 2026 flagship,
Mistral Medium 3.5
, merged chat, reasoning, and coding into a single 128B model, and their Robostral Navigate marked a first move into embodied robotics AI.
Mistral Medium 3.5
Mistral Large 2
Mistral Small 4
+2
Cohere
4 models
Cohere is an enterprise AI company focused on
large language models
for business applications, specializing in
RAG
, enterprise search, and sovereign AI deployments. Their May 2026 flagship,
Command A+
, is an open-source
MoE
model that runs on as little as two H100 GPUs.
Command A+
Command R+
Command R
+1
Midjourney
2 models
Midjourney
is an independent, self-funded research lab behind one of the most popular AI image generators. Famous for its distinctive aesthetic sensibility and a community that grew up inside Discord, it now runs primarily on the web with deep personalization — the model learns your taste from your ratings. It has never released
weights
or an
API
.
Midjourney
Midjourney V5
Black Forest Labs
2 models
Black Forest Labs is a German AI lab founded in 2024 by researchers behind the original
Stable Diffusion
, including Robin Rombach. Its FLUX family of image models spans fully managed
API
tiers down to Apache-licensed open
weights
that run on consumer GPUs — a deliberate open-to-closed spectrum that made FLUX the default choice for many production image pipelines.
FLUX.2
FLUX.1
Stability AI
2 models
Stability AI popularized open-weight image generation by releasing
Stable Diffusion
in 2022 — the moment AI image tools escaped the lab and landed on consumer GPUs. After leadership turmoil and financial trouble in 2024, the company refocused; its
Stable Diffusion
3.5 models remain a backbone of the local-generation ecosystem, powering community tools like ComfyUI and thousands of fine-tuned variants.
Stable Diffusion
Stable Diffusion 1.5
ElevenLabs
1 model
ElevenLabs is the leading AI voice company, offering text-to-speech, voice cloning, dubbing, sound effects, and music generation. Its models power audiobooks, game characters, accessibility tools, and countless voice agents. The flagship
Eleven
v3 tops blind "humanness" tests — listeners often cannot distinguish it from a real person.
Eleven
Suno
1 model
Suno
is the leading AI music generation company — type a description or lyrics and it produces a full song with vocals, arrangement, and production. It raised $400M at a $5.4B valuation in June 2026 while simultaneously settling and litigating with major record labels, making it the center of the AI-and-music-copyright fight. Its Studio product has grown into a full AI-native digital audio workstation.
Suno
Runway
2 models
Runway is an AI research company focused on tools for filmmakers and creative professionals. It co-created the original
Stable Diffusion
research, and its Gen series of video models pioneered commercial AI video. Runway tools have been used in Oscar-winning films and network TV production pipelines.
Gen-4.5
Gen-2
Perplexity
1 model
Perplexity
built its reputation on AI-powered answer-engine search with citations, then expanded into
Comet
, an AI-native web browser. It is a clear example of the "product layer" of AI: rather than training frontier models, it orchestrates models from OpenAI, Anthropic, and its own fine-tuned Sonar family into consumer tools.
Comet
Cursor
1 model
Cursor Agent
combines a selected model with instructions and tools for searching a codebase, editing files and running commands. Cursor documents supported hosted providers and bring-your-own
API
keys, with requests routed through its backend. These docs do not establish general support for an arbitrary localhost model server in Agent.
Cursor Agent
Aider
1 model
Aider
is an open-source terminal coding assistant with repository maps and git integration. It edits files with a selected model and normally commits its changes; automatic commits can be disabled.
Aider
documents both hosted models and local
Ollama
connections. Its workflow also supports editor integration and testing.
Aider
Cline
1 model
Cline
is an Apache-2.0 coding agent with VS Code and JetBrains integrations and a CLI. It can edit files, run commands and use tools with human approval. Local-model documentation includes
Ollama
and
LM Studio
. Auto-approve settings change which actions require confirmation.
Cline
OpenHands
1 model
OpenHands
provides an open-source agent platform and Agent Canvas for running coding agents and
automations
. Canvas supports local, remote and cloud backends and can connect other compatible agents. Local-model setup is documented. Runtime isolation depends on the chosen backend; running locally is not automatically a sandbox.
OpenHands
LM Studio
1 model
LM Studio
loads supported local models through
llama.cpp
and, on Apple Silicon, MLX. It offers chat and OpenAI-compatible
API
endpoints, plus headless operation through llmster.
LM Studio
can also host
MCP
tools in chat, so calling it
inference
-only is inaccurate. Offline operation applies to downloaded models and local features, not every integration.
LM Studio
Ollama
1 model
Ollama
manages models and serves
inference
through a CLI and
API
. The local server binds to 127.0.0.1:11434 by default and implements parts of the OpenAI
API
. It supports local models and optional cloud features; local-only mode can disable the cloud features.
Quantization
depends on the selected model tag or import configuration.
Ollama
llama.cpp
1 model
llama.cpp
is a C/C++
inference
project with
GGUF
model support and multiple hardware backends. It includes command-line tools and
llama
-server, which provides
API
serving and a web interface. Supported architectures and builds matter: a
GGUF
filename alone does not guarantee that a model can run.
llama.cpp
Learn
Companies
Fact Check
Compare
Sources