AI Models 1.1.0 Folds LLM Pricing, API Addresses, and SOTA Timelines into One Directory
AI Models 1.1.0 Released: Check LLM Prices, Compare Models, and Find API Base URLs
Swapping a large model into an app usually means doing two rounds of research. First, check the context window and the input/output costs, then go back to the vendor docs for the model ID and Base URL, and confirm which region and plan an API Key belongs to. Models that look the same can differ in access addresses and prices.
AI Models 1.1.0 puts both steps into a single directory. You can filter models, compare prices, then look up the API addresses a client needs. The new release also adds a SOTA timeline for original flagship models, making it easier to track new models from each vendor by release date. The November 15 version is built on Pi data, supplemented by official vendor sources, with 1,806 model records and 47 provider entries. One model may be offered by multiple platforms, so check the corresponding entry when reviewing prices. Open AI Models
Narrow Candidates with Combined Search
Enter DeepSeek 开源 on the home page and records matching both words appear; swapping the order also works. Search covers model names, model IDs, providers, and Chinese/English tags. To find image generation models, try Seedream 生图, or search for "free", "domestic", or "SOTA".
Search results can stack additional filters: multi-select provider, context range, flags for flagship, open source, and domestic. "Open source" marks whether weights are public; commercial terms depend on the model license. "Domestic" is judged by developer. Records whose open-source or domestic status is not yet verified can be isolated with the "unknown" filter.
The new version groups entries for a vendor's regions and plans together. After starring a frequently used entry, that vendor group stays at the top of the picker. As the table scrolls down, the current provider remains visible. To share a candidate set with colleagues, share a directory link with provider parameters, and the receiver opens the same set of providers.
Use Model PK to Calculate a Call Cost
On desktop, filter to the DeepSeek official entry, click "Start PK", select V4 Pro and V4.1 Flash, then click the compare button at the bottom. PK supports cross-provider selection; after switching filters, checked candidates stay selected.
Both models have roughly a 1.05 million token context, but the cost for the same request volume differs significantly. Using the dollar prices shown in the screenshot, assume one request with 100k input tokens and 10k output tokens, all input billed as standard input.
| 模型 | 输入/百万 tokens | 输出/百万 tokens | 单次估算 |
|---|---|---|---|
| DeepSeek V4 Pro | $1.30 | $4.00 | $0.17 |
| DeepSeek V4.1 Flash | $0.30 | $1.20 | $0.042 |
For actual selection, replace the sample input/output volumes with your own and try both models against the same real requests.
- Cache billing. The directory lists separate columns for input, output, cache read, and cache write. Click a price column header to sort within each provider group. A dash in cache write means the directory lacks a separate unit price.
- Currency conversion. Dollar is the default, but RMB, HKD, EUR, JPY, and KRW are available. The page displays the exchange rate date and supports manual refresh.
- Free and image generation. "Free" requires input and output unit prices to be explicitly zero, and recorded cache and image prices also zero. Image generation is quoted per token; estimating a single image's cost depends on actual consumption.
Find the Right Base URL by Protocol and Plan
If using Volcano Engine Ark, click "API Address" at the top, search volcengine, and you'll see pay-as-you-go API, Coding Plan, and Agent Plan. For Chat Completions, the three entries share a domain but use /api/v3, /api/coding/v3, and /api/plan/v3 respectively. When configuring a client, first confirm which entry the API Key corresponds to, then copy that row.
The address window supports multi-select providers, fuzzy search by name or directory ID, and protocol filtering. Pick the protocol matching your client and copy the Base URL on the corresponding row. Four protocols are supported: OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Google Generative AI. Each provider's support range is visible in the window. Braces in an address mark values to replace; each address links to its source for verifying authentication, region, and available models.
Zhipu is also grouped by access method. Z.ai international and BigModel China each differentiate pay-as-you-go API from Coding Plan. Ark and Zhipu plans are quota-based; the directory writes the billing method in the entry's description, and the token unit price column is left blank.
View Original Flagships on the SOTA Timeline
From the home page, filter for SOTA to find flagged flagships in the current candidates. To see the latest releases from each vendor, open the standalone timeline. Cards are sorted by release date descending, listing vendor, model ID, capability tags (general or reasoning), and official references. The timeline shows only original flagships; region and plan entries for the same model are merged. View SOTA model page
Here SOTA is a manually verified vendor flagship marker, covering general, reasoning, coding, and image directions. Click a model marker in the directory to see the verification date and source. Unmarked entries may still be under review.
Start with a Model You're Already Using
On mobile, use the horizontally scrollable list for quick model and price lookup. On desktop, search the model you're using, add alternatives to PK, and calculate against your own usage. Once a model is selected, go to "API Address" and copy the Base URL for the matching protocol and plan.
Filter, sort, currency, favorites, language, and theme preferences are saved in the current browser. To start a fresh set, click "Reset" to restore default filters and price display.
For details on price columns, version changes, and data methodology, check these resources. Project intro · Changelog · Source & data methodology