Last updated: 2026-10-13
APIMODELS is an AI model gateway whose pricing page lists more than 120 image, video, language and audio models behind one API key. It accepts OpenAI-style and Anthropic-style requests at api.apimodels.app, bills only successful calls, and ships an MCP server that lets coding agents render images, video and speech.
About APIMODELS
APIMODELS (apimodels.app) is a gateway that puts image, video, language and audio models from Anthropic, OpenAI, Google, ByteDance, MiniMax and other makers behind a single API key and a single USD balance. The site names no founders and no operating company. Its terms and privacy policy are both dated 2026-06-06, the same day its public examples repository (apimodels/apimodels-api-demo, MIT licence) was created, which is the earliest dated evidence of the service that turned up in research.
Developers point the OpenAI SDK at https://api.apimodels.app/v1, or the Anthropic SDK at the root host, and change one model string to move between listed language models such as Claude Opus 5.5 and GPT-6 Sol, with models from xAI and DeepSeek in the same list. Image, video and audio models use their own endpoints, and long video jobs run asynchronously with a task id, a status check and webhook callbacks. The video list includes Seedance 2.5, Wan 3.0, MiniMax H3 and VEO 3.1, with Kling clips alongside them, and the audio list carries ElevenLabs voices plus Suno music.
Treat it as a price-and-convenience layer rather than a compliance layer; the cons list the gaps worth checking before any production traffic moves over, and a small paid test is the sensible first step.
For coding agents there is an MCP server, apimodels-mcp (MIT licence, version 0.7.0 on npm, also hosted at https://api.apimodels.app/mcp), which lets coding agents such as Claude Code, and editors such as Cursor, generate and review images, video and speech from inside the chat. Among gateways already reviewed on HokAI, OpenRouter leans toward text routing and has a funded track record. For media generation, AIML API has the wider catalogue and a higher prepaid minimum, while CometAPI offers a trial credit without a card.
Pricing
Pay as you go in US dollars with no monthly fee; the minimum top-up is $10 and failed requests are not billed. 01 per second (MiniMax H3 Lite, 480p). The terms name Stripe and Alipay for payment, and the homepage FAQ also lists PayPal and USDT.
Prices follow upstream costs and can change, so the model page is the figure that bills. app.
| Tier | Monthly price | What it includes |
|---|---|---|
| Pay-as-you-go | Per request | Prepaid USD credits, no monthly fee, failed requests not billed, every listed model on one key |
| Business | Contact sales | Partnerships, enterprise volume and custom needs via business@apimodels.app |
Key Features
- One key across four modalities: The pricing page groups its catalogue into image, video, language and audio models, so one key and one USD balance cover Claude, GPT, Seedance, Kling and ElevenLabs calls.
- Two SDK-compatible endpoints: api.apimodels.app/v1 serves OpenAI-style /v1/chat/completions and Anthropic-style /v1/messages, and the docs recommend it because it accepts request bodies up to 100MB against about 4.5MB on the alternate host.
- Success-only billing with typed failures: A failed async task returns one of 8 standard failCode values plus a retryable flag, and the terms say failed requests are not charged.
- Multi-channel failover: The vendor says popular models run on more than one upstream channel, and an errored or rate-limited request is retried on a backup channel with no client change.
- Per-request records: Any request can be looked up with GET /v1/records/{id}, and the console has Records and Usage views for reconciling spend.
- MCP server for coding agents: apimodels-mcp exposes generate_image, generate_video, review_image, chat and get_balance tools, and the hosted version needs only a sign-in and one Authorize click.
Pros
- Published per-model prices make the discount checkable: the Anthropic list price it cites ($4 / $20 per million tokens) matches HokAI's own model record.
- The entry cost is low next to prepaid gateways such as AIML API, whose minimum top-up is $20, so a trial costs little.
- The MCP server and the API examples repository are MIT-licensed on GitHub, so the integration code can be read before a key is created.
Cons
- No operating company, founder or business address is named anywhere in the site documents, and no independent review turned up, so counterparty risk on a prepaid balance is hard to judge.
- Purchased credits are generally non-refundable under the terms, so a top-up is a sunk cost if the service changes or disappears.
- Generated media is kept on Cloudflare R2 for only a month before deletion, so anything worth keeping must be copied out.
- The homepage counters disagree: a 79% maximum saving sits beside an up-to-95% headline, the savings table lists 10% to 63% on seven models, and an uptime counter read 70.4% with no stated window. No public status page was found at /status.
- No SOC 2, ISO 27001, HIPAA or GDPR statement is published; prompts and uploads go to third-party providers, and the privacy policy allows processing in several countries.
- Some listed models carry their own warnings, for example one Seedance Global variant is marked testing and personal studio use only and may fail with a refund.
Data Handling
- Training-data policy
- Not stated. The privacy policy says inputs are forwarded to third-party model providers and does not describe any training use.
- Data retention
- 30 days
Frequently Asked Questions
How much do you pay for APIMODELS?
It is pay-as-you-go from a prepaid USD balance with no subscription, and each model page shows its own price. Examples from the pricing page are GPT-6 Sol at $1 per million input tokens and $5 output, Dreamina Seedance 2.0 Mini from $0.0225 per second, and GPT Image 2.5 Flare from $0.008 per image. The homepage savings table lists 10% to 63% off official prices on seven models, checked by the vendor on 2026-09-30, and per-second video is billed at the maximum requested duration.
Is APIMODELS free to use?
Creating an account needs no identity verification or payment method, but the pages checked do not mention free trial credit, so assume paid calls begin after the first top-up. There is no monthly fee. Failed requests are not billed, which limits waste while you test.
What are the best alternatives to APIMODELS?
OpenRouter suits teams that want a funded, longer-running gateway with a very large model list. CometAPI is the one to test if a trial credit with no card matters. AIML API fits buyers who need the widest modality list, 3D included, and accept a higher prepaid minimum. Replicate is the route when you run open-source models billed by GPU time.
APIMODELS or OpenRouter: which should you pick?
Pick OpenRouter when the workload is mostly text, you want provider pricing with a disclosed purchase fee, and a documented funding history matters. Reach for APIMODELS if image and video generation is the main spend, or if coding agents need an MCP server. Neither is a compliance shortcut, since both forward prompts to upstream providers.
What does it take to start using APIMODELS?
Sign up with Google, Alipay or email, create a key in the console, then set the base URL to https://api.apimodels.app/v1 in an OpenAI-compatible SDK, or to the root host for the Anthropic SDK. A minimum top-up is needed before paid calls, after which switching models means changing one model string. Coding-agent users can instead run npx -y apimodels-mcp login and install the plugin.
Top Alternatives
- OpenRouter: Pick OpenRouter for a funded, documented track record and a very large model list; pick APIMODELS when image and video generators drive the bill.
- CometAPI: Pick CometAPI if a no-card trial credit and a larger catalogue matter; pick APIMODELS when an MCP server for coding agents does.
- AIML API: Pick AIML API for the widest modality list including 3D; pick APIMODELS for a lower minimum top-up.
- Replicate: Pick Replicate to run open-source models billed by GPU time; pick APIMODELS for flagship closed models billed per call.