Skip to content
Draft
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion v3/llms/listing-models.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -92,10 +92,11 @@
"regions": [
{ "code": "eu", "name": "Europe" }
],
"zero_data_retention": false,

Check warning on line 95 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L95

Did you really mean 'zero_data_retention'?
"quantization": null,
"retirement_date": null,

Check warning on line 97 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L97

Did you really mean 'retirement_date'?
"discount_ends_at": null,

Check warning on line 98 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L98

Did you really mean 'discount_ends_at'?
"alias_of": "google/gemini-3.6-flash"

Check warning on line 99 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L99

Did you really mean 'alias_of'?
}
]
}
Expand All @@ -103,7 +104,7 @@

Each model `id` is used directly as the `model` parameter in your requests.

`pricing` holds the rates a call bills at, after any discount, and `list_pricing` the provider's list rates before it, most of them in US dollars per token. Models that can run as a [batch](/v3/llms/batches) (OpenAI, Anthropic and Mistral) also carry `input_cost_per_token_batches`, `output_cost_per_token_batches` and `cache_read_input_token_cost_batches`, the rates a batch bills at when the model has them.
`pricing` holds the rates a call bills at, after any discount, and `list_pricing` the provider's list rates before it, most of them in US dollars per token. When a model is discounted, `discount` gives the fraction off (in the `0` to `1` range) and `discount_ends_at` gives the UTC ISO 8601 timestamp when the discount ends and the model returns to `list_pricing` (`null` when there is no discount or no announced end). Models that can run as a [batch](/v3/llms/batches) (OpenAI, Anthropic and Mistral) also carry `input_cost_per_token_batches`, `output_cost_per_token_batches` and `cache_read_input_token_cost_batches`, the rates a batch bills at when the model has them.

## Grouped view: one entry per routable name

Expand Down Expand Up @@ -149,7 +150,7 @@

What to expect from the shape:

- **`id` is the routable name.** Send it as `model` and Eden AI picks which provider serves it. Rerank is the exception: `POST /v3/rerank` refuses a name without its provider with `400`, so from `/v3/rerank/models` send one of the `endpoints[].id` values instead.

Check warning on line 153 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L153

Did you really mean 'routable'?

Check warning on line 153 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L153

Did you really mean 'Rerank'?
- **Top-level fields summarise the model across its endpoints.** `context_length` is the largest window among the endpoints. `capabilities` is the union across endpoints, so a flag is true and a modality listed when any endpoint supports it. `pricing` carries the full price list of the cheapest endpoint for a typical call (untagged endpoints only, since an `@region` variant prices a tagged call), identified by `pricing_endpoint`; among endpoints at the same input and output prices, it picks the one with the cheapest cache reads, then cache writes (an endpoint that states no cache rate comes after those that do), then the model's own lab when it is one of them. Both are `null` when no endpoint has a price. A routed call is billed at the prices of the provider that served it. `regions` lists every region where at least one endpoint runs, sorted by code.
- **Each entry in `endpoints` is the same object the default view returns**, with that provider's own pricing, capabilities, context length, and regions. To compare providers or see per-provider values, read `endpoints`.
- **`owned_by` changes meaning between the views.** On a grouped entry it is who authored the model (`openai`); on an endpoint entry it is the provider selling it (`azure`).
Expand Down Expand Up @@ -203,11 +204,11 @@
| `fp8`, `int8` | 8 bits per weight |
| `fp6` | 6 bits per weight |
| `fp4`, `int4` | 4 bits per weight |
| `null` | No precision to show: the provider does not state one, or an Ollama level has no value in this list |

Check warning on line 207 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L207

Did you really mean 'Ollama'?

The bit count is nominal. Some formats keep a few tensors at a higher precision and store scaling data alongside the weights, so read it as a category rather than an exact memory size. The value is what the provider states; Eden AI does not measure it. It is `null` for every closed-weight model (OpenAI, Anthropic, Google Gemini and similar), whose makers do not publish how they serve them. In the default view, an alias with no provider shows a precision only when every provider behind it serves the same one.

For a model on your own private deployment, the value comes from the level your Ollama server reports: `Q4_0`, `Q4_1`, `Q4_K_S` and `Q4_K_M` are `int4`, `Q8_0` is `int8`, and `F16`, `BF16` and `F32` are `fp16`, `bf16` and `fp32`. Any other level is `null`, and the model's `description` still names it.

Check warning on line 211 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L211

Did you really mean 'Ollama'?

## Retirement date

Expand All @@ -229,9 +230,9 @@
| Alias | `alias_of` | Who serves the call | Listed in |
|-------|-----------|---------------------|-----------|
| With a provider, such as `anthropic/claude-sonnet-latest` | A `provider/model` id | That provider, unless the call falls back to a model you listed in [`fallbacks`](/v3/general/fallback) | The default view |
| Without a provider, such as `~claude-sonnet-latest` | A routable model name | The provider Eden AI picks, as with [Provider Routing](/v3/llms/provider-routing) | `view=models` |

Check warning on line 233 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L233

Did you really mean 'routable'?

An alias without a provider starts with `~` (e.g. `~claude-sonnet-latest`) so it can never share a name with a model. In a shell, quote it (`--model '~claude-sonnet-latest'`): some shells, such as zsh, expand an unquoted leading `~`. It is listed with `view=models` only, where it repeats the entry of the model it names, endpoints included, as described under the [grouped view](#grouped-view-one-entry-per-routable-name). A call on it is billed at the prices of the provider that served it.

Check warning on line 235 in v3/llms/listing-models.mdx

View check run for this annotation

Mintlify / Mintlify Validation (edenai) - vale-spellcheck

v3/llms/listing-models.mdx#L235

Did you really mean 'zsh'?

The catalog lists both forms when a provider exposes them, so you may see a stable alias (e.g. `anthropic/claude-opus-4-5`) and one or more dated snapshots (e.g. `anthropic/claude-opus-4-5-20251101`) side by side. Use the stable alias to always get the latest version, or a dated snapshot to pin to a specific release.

Expand Down
Loading