Comparing models

Search the catalogue

Open Model catalogue on the public website or in the workspace. Search by model name, source identifier, or training house. On the public site you can also filter the training house and input modality.

The catalogue retains the available public source records. A search filter narrows what is shown; it does not redefine which suppliers are commercially onboarded.

The public website also keeps a Provider directory of the provider records retained from the same source. Like the model catalogue, these are dated reference listings for research, not proof of onboarded supply or available capacity.

Where the catalogue comes from

The Megatron Markets catalogue is a reference library aggregated from dated public catalogue sources. Each record keeps its original published fields and its retrieval date, so you can see exactly what was published and when. Megatron Markets does not write the models and does not claim authorship of the records: the training house named on each record is the model's author, and the catalogue simply preserves what was publicly listed.

Everything you need for research is on this site. The Model catalogue and Provider directory carry the retained descriptions, context lengths, input and output prices in $ / 1M tokens, modalities, supported generation settings, and endpoint observations, so there is no need to consult an outside listing. If a detail you need is not shown, it was not published in the source record — ask us through Contact Us and the team will help you confirm it with a supplier.

Read a model row

Each row shows the model name, model identifier, context window, and separate input and output reference prices in $ / 1M tokens. Open the model to inspect its full record.

Context length is the amount of information a model can handle within a request. It is different from a model’s parameter count and from your purchased allowance.

Review the details

Use the detail tabs to inspect:

• Specifications: training house, model identity, input/output modalities, context size, output limit, tokenizer, architecture fields, supported generation parameters, and defaults.
• Prices & providers: exact published source values, comparable $ / 1M tokens rates only for token-priced fields, conditional overrides, provider endpoint observations, and their retrieval dates. A source field without a declared unit is labeled as such rather than converted.

Expand the detailed sections when you need a field that is not in the main summary. Unknown or unpublished information is labeled Not published. Do not interpret a missing value as zero.

Retrieval dates identify when the platform fetched the published information. A failed import does not confirm supplier availability or terms. The model and provider pages show product information and price schedules without raw source-record panels.

Compare the right things

Two endpoints serving the same model may differ in price, context limits, region, retention, throughput, or execution mode. A model-level headline price is not a promise about every endpoint.

Supported generation parameters, such as temperature or a reasoning setting, describe request controls. Architecture metadata, such as parameter count or tokenizer, describes the model. These are separate categories.

A verified Hugging Face mapping may provide additional metadata. A source identifier alone is not proof that every repository field has been verified.

From research to a quote

Select Request a quote on a model to open a prefilled RFQ in the workspace. Review the selected model and add your quantity, delivery window, and constraints before submitting. Alternatively, choose Enquire about this model to speak with the team without an account.

Published modalities describe the model source. Your accepted endpoint and the configured gateway must also support a capability before it can be used. See API access.