The most capable model in the catalogue. For long-horizon agentic work where being right matters more than being cheap.
The catalogue,
laid out plainly.
Over a hundred models sit behind the key. These are the ones worth knowing by name, arranged by what each is for. Read down a column for one provider's ladder, across a row to compare the same rung between providers.
Anthropic
OpenAI
Gemini
DeepSeek
MiniMax
FlagshipWhen being wrong is expensive
OpenAI's flagship tier. For the hardest problems, from complex coding to research, where being wrong is expensive.
Google's top tier. Long context and native multimodality for the work that needs both at once.
DeepReasoning and long agentic work
Deep reasoning at a little over half the price of Fable. The default for complex agentic coding and analysis.
$4.75 in, $23.75 out, per 1M tokensOpen-weight reasoning with a built-in thinking mode, at a fraction of frontier pricing.
Open-weight, frontier coding, long context and native multimodality in one model.
BalancedThe everyday workhorse
Near-Opus quality on coding and agentic work at a fraction of the cost. The everyday workhorse.
$2.85 in, $14.25 out, per 1M tokensThe balanced everyday tier. Built for high-volume business work like support, internal tools and document analysis.
Reads images, audio and long documents natively. Strong on coding and agent workflows.
FastHigh volume, low cost
Built for volume. Classification, extraction, and anything you run a million times a day.
The cost-efficient tier. Good for summarising, drafting and routine automation at scale.
Low latency and very cheap. Built for high-volume automation and subagent work.
One of the cheapest capable models available. Hard to beat on cost per token.
OutsideNot on the ladder at all
Text and image to video, with strong Chinese language performance. The model behind our own campaign footage.
Turns a still or a written prompt into video. Priced per second of footage, not per token.
Run open models on your own hardware. Nothing leaves your building.
- Text
- Multimodal
- Video
- Local
Prices are per million tokens, or per second for video. Current pricing for every provider lives on the Hub, because it changes more often than this page does.
How to choose
A crucible is judged by what comes out of it, not by what it cost.
Match the model to the job, not to the hype.
Most work does not need a frontier model. Classification, extraction and routine replies run happily on a fast model at a fraction of the cost. The table above puts the price on every tile for exactly that reason: GPT-5.4 mini costs a third of the model above it. The point of one key is that you can make that call per task instead of per contract.
Switching
Change your mind without changing your plumbing.
Models move fast. Something cheaper and better ships every few months. Because every model here sits behind the same key and the same interface, switching is a configuration change, not a migration project. That is most of the value.
The elements are all here. The combination is yours.
Tell us what you sell. We will tell you what to automate first.
Monday to Friday, 9:30am to 6:00pm