Model catalogue

Latest frontier agentic coding model — the new Codex default.

OpenAI's flagship for coding, agents and research-grade reasoning.

The fast, light GPT-5.6 — cheap agents and high-volume coding.

Balanced agentic coding model for everyday work.

Production workhorse — near-flagship quality at half the price.

The strongest mini — cheap sub-agents and high-volume tasks.

OpenAI's open 120B model at medium reasoning depth

OpenAI's smaller open model — high-volume work at no charge

OpenAI's previous image generation — follows the prompt closely

OpenAI's cheap speech — bot replies and notifications

OpenAI images: generation and edits from a description

OpenAI images, a generation newer than the first

OpenAI's latest images — text in frame and prompt-based edits

OpenAI vectors for semantic search — the larger of the two

OpenAI vectors at a lower rate — big corpora and frequent reindexing

Legacy OpenAI vectors — compatibility with older indexes

Fast OpenAI speech — streaming with little delay

A step up in OpenAI speech — recordings and voice-over

The newest experimental Opus, served through Kiro.

The current Opus — Anthropic's agentic-coding flagship.

The new mid-tier Claude 5 — intro-priced by the vendor until September.

Previous Opus — between 4.8 and 4.6 in the metering ladder.

Previous-generation Opus — flagship quality, smaller multiplier.

The balanced Claude — fast, smart, affordable.

Cheapest, fastest Claude for high-volume tasks.

Anthropic's Mythos-class tier above Opus — #1 on our AI Arena.

Previous-generation flagship Claude — long analysis and code

Previous-generation workhorse Claude — balanced price and quality

Moonshot's multimodal flagship for coding, games and knowledge work.

Moonshot's strongest coder, always-thinking.

Zhipu's MIT-licensed flagship — a coding-agent favourite.

GLM with vision: images, screenshots and documents

Previous-generation GLM — inexpensive everyday text

Fifth-generation GLM — code, agents, long tasks

GLM 5.1 — the same generation, a step above the base five

The top GLM of the line — the family's heaviest tasks

xAI's current frontier model — its own docs' default pick for code.

Grok's coordinated multi-agent mode for deep research.

Grok with reasoning on — multi-step problems

Grok 4.3 — an earlier generation at a lower rate than the top models

Grok 4.5 — the line's general-purpose pick

Grok tuned for building code — project edits and agent runs

The same Grok without reasoning — fast short answers

xAI images — fast generation from a description

The same xAI images in quality mode — slower per frame

xAI video — short clips from a description

xAI video, a generation newer

The cheapest serious reasoner on the market.

Open-weight frontier reasoning at rock-bottom prices.

Google's newest default — frontier smarts at a Flash price.

The fastest, cheapest Gemini of the 3rd series.

Google's open-weight model — near-free tokens for bulk work.

Fast Gemini 3 — cheap request volume and simple agents

Gemini 3.6 Flash at high reasoning — multi-step problems

Gemini 3.7 Flash at high reasoning — harder analysis

Gemini Pro for agents — long tool-using runs

An open 26B Gemma built on diffusion

Gemini 2.5 Flash — earlier generation, cheap everyday text

The lightest Gemini 2.5 — simple tasks at volume

Gemini 2.5 Pro — long context and document analysis

The same Gemini 3 Flash tuned for tool-using agent runs

Gemini images — generate and edit a frame from a description

Gemini 3.1 Pro at low reasoning — cheaper than the full run

Gemini 3.5 Flash at minimal reasoning — the cheapest run

Gemini 3.5 Flash at low reasoning — quick answers

Gemini 3.6 Flash, low reasoning — request volume

Gemini 3.6 Flash at medium reasoning — the working middle

Latest Gemini 3.7 Flash, low reasoning — cheap volume

Gemini 3.7 Flash at medium reasoning — everyday work

Imagen 3 — Google's photoreal generation

Google's prompt-based image editing — fast and cheap

Lighter image editing — drafts and bulk runs

Google's top image editing — clean detail and text

Google video — a scene generated from a description

Veo 3.1 — video from a description, with sound

Lighter Veo — draft clips at lower cost

Long-context multimodal reasoning at a budget price.

NVIDIA's open reasoning flagship, available free through AnyModel.

A small vision Llama from NVIDIA — images and documents

Mistral, tuned by NVIDIA — everyday text and tools

Nemotron 3 Nano — a light 30B model at no charge

Nemotron Nano Omni with reasoning — text, images and audio together

The top 120B Nemotron — the line's heaviest tasks

NVIDIA content safety — flags disallowed material in text

The line's fastest Nemotron — short answers at speed

A 12B vision Nemotron — screenshots and tables

The lightest 9B Nemotron — bulk runs at no charge

NVIDIA Riva translator — streaming text translation

NVIDIA vectors built on Llama — search over your own corpus

NVIDIA vision vectors on the gateway's second rate

NVIDIA vision vectors — search across images and documents

1B Nemotron vectors — text indexing at no charge

Speech transcription on Nemotron Omni — audio to text

NVIDIA's classic vectors — general semantic search

NVIDIA vectors for code — search across a repository

NVIDIA vectors for question answering — retrieval for RAG

Efficient Qwen MoE model with low input cost.

8B Qwen3 vectors — multilingual semantic search

Deepgram transcription — streaming speech to text

Nova 2 — a generation newer than the first

Third-generation Nova — Deepgram's current transcription

Whisper Large via Deepgram — multilingual transcription

FLUX.1 Dev — the open research build for images

FLUX.1 Schnell — fast generation in a few steps

4B FLUX.2 Klein — the smallest open model of the line

The smaller Perplexity vectors — inexpensive indexing

4B Perplexity vectors — the larger of the pair

Automatically selects a currently available free model.

A small gateway text model — simple tasks at no charge

A compact Laguna model — short answers at no charge

11B vision Llama 3.2 — an open model for images

Step 3.7 Flash — a fast model for request volume