staticdata.dev
Menu
Datasets About Contact

LLM Models

A snapshot as of October 2026 of production language model metadata (45 models) across the major providers — Anthropic, OpenAI, Google, Mistral, DeepSeek, xAI, Meta, Alibaba, Cohere — with each model's context window, max output tokens, input and output prices in USD per million tokens, vision and tool-use capabilities, knowledge cutoff date, release date, and deprecation date if announced.

v1.0.0 MIT 45 records Updated 2026-10-05 stable Source:  Anthropic models overview, Anthropic pricing, Anthropic model deprecations, OpenAI models, Google Gemini models

Quick fetch

Stable URL. CORS-enabled. No auth.

bash 
curl https://staticdata.dev/v1/llm-models.json

Or use as a typed import: import { llmModels } from "https://staticdata.dev/v1/llm-models.ts"

Formats

  • JSON 19.3 KB

    /v1/llm-models.json

    Open

    min: 15.6 KB · /v1/llm-models.min.json

  • CSV 6.3 KB

    /v1/llm-models.csv

    Open
  • TypeScript 19.5 KB

    /v1/llm-models.ts

    export const llmModels

    type LlmModel = (typeof llmModels)[number]

    Open

Schema

Each record in the dataset has the following shape.

Schema
Field Type Description Example
id string API identifier exactly as accepted by the provider claude-opus-4-7
provider string Provider name Anthropic
family string Model family the version belongs to Claude 4
contextWindow number Maximum context window in tokens (input + output) 1000000
maxOutputTokens number Maximum output tokens per synchronous request (set equal to the context window where the provider publishes no separate output cap) 128000
inputPricePerMillionUsd number Standard input price per million tokens in USD 5
outputPricePerMillionUsd number Standard output price per million tokens in USD 25
supportsVision boolean Whether the model accepts image inputs true
supportsTools boolean Whether the model supports tool / function calling true
knowledgeCutoff string ISO YYYY-MM month of the model's (reliable) knowledge cutoff, or "unknown" where the provider publishes none 2026-01
released string ISO YYYY-MM month the model was released 2026-04
deprecated? string | null ISO YYYY-MM month the provider announced the model's deprecation (retirement notice), or null while it is active
sourceUrl string URL to the provider's documentation for the entry https://platform.claude.com/docs/en/about-claude/models/overview

Preview

First 10 records.

Data preview
idproviderfamilycontextWindowmaxOutputTokensinputPricePerMillionUsd
claude-opus-4-7AnthropicClaude 410000001280005
claude-sonnet-4-6AnthropicClaude 410000001280003
claude-haiku-4-5AnthropicClaude 4200000640001
claude-opus-4-5AnthropicClaude 4200000640005
claude-3-5-sonnet-20241022AnthropicClaude 3.520000081923
claude-fable-5-1AnthropicClaude 5100000012800010
claude-opus-5-5AnthropicClaude 510000001280004
claude-sonnet-5-5AnthropicClaude 510000001280002
claude-fable-5AnthropicClaude 5100000012800010
claude-opus-5AnthropicClaude 510000001280005

Showing 10 of 45. View full data:  JSON · CSV

Fetch examples

Drop-in snippets in five languages.

curl 
curl -sSL https://staticdata.dev/v1/llm-models.json | jq '.[0]'
JavaScript 
import type { LlmModel } from "https://staticdata.dev/v1/llm-models.ts";

const res = await fetch("https://staticdata.dev/v1/llm-models.min.json");
if (!res.ok) throw new Error(`Fetch failed: ${res.status}`);
const llmModels: LlmModel[] = await res.json();
console.log(llmModels[0]);
Python 
import urllib.request, json

with urllib.request.urlopen("https://staticdata.dev/v1/llm-models.min.json") as r:
    llmModels = json.load(r)

print(llmModels[0])
Go 
package main

import (
	"encoding/json"
	"fmt"
	"net/http"
)

func main() {
	resp, err := http.Get("https://staticdata.dev/v1/llm-models.min.json")
	if err != nil { panic(err) }
	defer resp.Body.Close()

	var data []map[string]any
	if err := json.NewDecoder(resp.Body).Decode(&data); err != nil {
		panic(err)
	}
	fmt.Println(data[0])
}
Rust 
use serde_json::Value;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let body = ureq::get("https://staticdata.dev/v1/llm-models.min.json").call()?.into_string()?;
    let data: Vec<Value> = serde_json::from_str(&body)?;
    println!("{:?}", data.first());
    Ok(())
}

Sources and methodology

This dataset moves faster than any other on the site. Providers ship new models, change prices, deprecate old endpoints, and revise context-window limits on a near-monthly cadence. This is a snapshot as of October 2026: values were re-checked against each provider’s official model, pricing, and deprecation pages in early October 2026. Verify against the provider’s docs before making pricing or capability decisions in production.

Each entry’s sourceUrl points to the provider’s official documentation page. If a price or limit looks wrong, click through and check; we update from those pages.

Pricing is the standard, on-demand API price for the base tier (for example the lower prompt-length tier where a provider charges more for very long prompts, and DeepSeek’s standard rather than off-peak price). Provisioned throughput, batch discounts, long-context surcharges, fine-tuned model surcharges, promotional prices, and prompt-caching discounts are not represented.

Context windows are the maximum token capacity per request. The maxOutputTokens is the per-request output cap on the synchronous API, which is often much smaller than the full context window; where a provider publishes no separate output cap (some Mistral, xAI, and hosted Llama listings), it is set equal to the context window. knowledgeCutoff is "unknown" where the provider does not publish a cutoff; for Anthropic models it is the “reliable knowledge cutoff” from the models overview.

Models from open-weight families (Llama, Qwen, DeepSeek, Mistral) have prices reflecting hosted-API pricing from the model’s primary provider (Llama 3.3 70B: Together AI serverless); running these models on third-party infrastructure or self-hosted will cost differently.

The deprecated field holds the month the provider announced the model’s deprecation or retirement; null means no deprecation has been announced as of October 2026. Retired models (for example claude-3-5-sonnet-20241022, retired in October 2025, or gemini-2.0-flash, shut down in June 2026) are kept with their deprecated month instead of being removed so existing ids stay stable; check the provider’s deprecation page for the exact shutdown date. Price and limit values for retired models are their last published values.

Versioning

URLs under /v1/ are stable: field names and types will not change in a way that breaks consumers. Values can be corrected or records added when the upstream source changes; schema-incompatible updates ship under a new version path. See the llm-models changelog for this dataset's history.