Published Updated
In one or two sentences, before anything else.
Kie.ai is a credit-based API service, run by two Denver LLCs, that sells access to other companies' video, image, music and language models at prices well below the makers' own, such as Veo 3.1 Quality at $1.275 per 1080p clip against Google's $3.20. It does not say how it obtains that access.
What it can do, one sentence each, and whether we actually tested it.
590 price rows across 22 maker labels: Veo 3.1, Kling 3.0, Seedance 2, Runway, Nano Banana Pro, GPT Image 2, Flux, Suno, Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon on one key.
Kie.ai's pricing table; it names no partnership, licence or reseller status with these makers.
Veo 3.1 Quality at 1080p for $1.275 against Google's $3.20; Nano Banana Pro for $0.09 against $0.134; Claude Opus 5.5 at $1.60 / $8.00 per million tokens against Anthropic's $4 / $20.
Its own discount column is headed 'Official / Fal Price' and often measures against fal rather than the maker.
Credits cost $0.005 each (200 per dollar) on 582 of 590 rows; new accounts get 80 free credits and credits never expire.
The 200-per-dollar rate and the $0.40 value of the free credits are arithmetic from Kie.ai's pricing table.
Kie.ai lists 'Failed Tasks $0' and says pre-deducted credits for failed tasks come back automatically.
Trustpilot reviewers in June and October 2026 say credits were used on jobs stuck at 99%.
Image, video and music jobs return a task ID that you poll or receive by callback; the default limit is 20 new requests per 10 seconds, and excess requests get a 429 error.
Per Kie.ai's docs; support can raise the limit.
Anthropic Messages at api.kie.ai/anthropic, with a Claude Code setup guide, and the OpenAI Responses API at api.kie.ai/openai/v1, with a Codex CLI guide.
Kie.ai's AI agent docs also cover Grok Build.
kie-models lets a coding agent pick, run and download from 200+ media models; kie-chat-agents sets coding agents up on Kie.ai's chat models. Both install with npx skills add https://kie.ai.
Per Kie.ai's AI agent overview; no official MCP server was found in its docs.
The people it helps and the problems it solves, and where it ranks.
Built for
Problems it solves
Graded in
Public tests other people ran: what they gave it and how it did.
Case 1
Input
A text-to-video task for Veo 3.1 Fast at 720p, 8 seconds long, collected by callback or by polling the task ID.
What happened
Kie.ai charges 60 credits, $0.30, for a 4, 6 or 8-second clip, against $0.80 for 8 seconds at Google's rate and $1.20 on fal, Replicate and WaveSpeedAI. Files are deleted after 14 days, its Veo page calls audio experimental, and Trustpilot reviewers report jobs stuck at 99%.
Case 2
Input
Claude Code started with ANTHROPIC_BASE_URL=https://api.kie.ai/anthropic and ANTHROPIC_AUTH_TOKEN set to a Kie.ai key, choosing a model listed by GET /anthropic/v1/models.
What happened
Claude Opus 5.5 is listed at $1.60 / $8.00 per million tokens, 60% below Anthropic's $4 / $20. Kie.ai does not say how it gets Claude access, the guide does not say whether prompt caching or web search work, and no public test checks that answers come from Opus 5.5.
Case 3
Input
An agency's integration of Kie.ai's image and video endpoints, from sign-up through two weeks of use.
What happened
Onboarding took under 15 minutes and a support reply came about 10 hours after asking; video tasks failed intermittently and had to be retried by hand. The review gives no counts, and its claim that Kie.ai is not a licensed partner cites no source.
What it does badly, with the evidence, and what we could not settle.
Weak
It does not say how it gets access to the models.
Evidence: Kie.ai's homepage, About page and terms make no claim to be an authorised reseller or partner of any model maker; the terms say the trademarks shown do not imply endorsement. The About page names two LLCs while the footer reads 'kie.ai Inc.'.
Its savings figures overstate the discount against the makers.
Evidence: The discount column is headed 'Official / Fal Price'. For Veo 3.1 Fast at 720p it compares with $1.20, fal's price, not Google's $0.80, so the claimed 75% saving is 62.5% against Google (pricing table, October 21, 2026).
Not the cheapest on every model.
Evidence: Seedance 2.0 at 720p costs $0.205 a second on Kie.ai against $0.18 on Replicate (each pricing page, October 21, 2026).
Public reliability complaints and no status page.
Evidence: Trustpilot: 2.1 out of 5 from 16 reviews, 81% one star, with jobs stuck at 99% while credits were used (June and October 2026) and 8 to 10 hours of 500 errors (May 2026). There is no public status page, and the homepage claims both 99.9% uptime and 99% availability.
Few protections for paying customers or their data.
Evidence: Accounts can be suspended or closed at Kie.ai's sole discretion without notice; the refund policy gives no conditions or time window; the privacy policy covers only your email address and says nothing about prompts or uploads.
Not for confidential data or production systems without a fallback.
Evidence: With no stated sourcing, no data terms for prompts and outputs, discretionary account closure and public reports of hours-long errors, businesses with customer data or uptime needs should use the makers' own APIs or a provider that states its terms.
Uncertain
Whether outputs come from the named model and version.
Why unsure: No public comparison of Kie.ai's outputs with the makers' APIs was found, and its 'smart fallback routing' is not explained.
How long prices this low will last.
Why unsure: Claude Opus 5.5 at $1.60 / $8.00 is below Anthropic's own Batch rate of $2 / $10, and Kie.ai gives no reason it can sell under the maker's lowest price.
Minimum top-up and accepted payment methods.
Why unsure: The billing page needs a login; a March 2026 Trustpilot reviewer mentions $5 for 1,000 credits, which no Kie.ai page confirms.
How representative the complaints are.
Why unsure: Sixteen Trustpilot reviews are a small, unverified sample, and no news outlet or benchmark has measured Kie.ai.
The numbers, each with its source and the date we checked it.
| Spec | Value | Note and source |
|---|---|---|
| Operator | NEXUSAI SERVICES LLC and INNOLEAP AI LLC, 118 Krameria St, Denver, Colorado, US; terms effective August 1, 2025, Colorado law | The site footer reads '© 2026 kie.ai Inc.'. Source: kie.aichecked |
| Price against the maker's list | Veo 3.1 Quality 1080p $1.275 vs Google $3.20; Nano Banana Pro $0.09 vs $0.134; Claude Opus 5.5 $1.60 / $8.00 per million tokens vs Anthropic $4 / $20 | Kie.ai's own discount column is headed 'Official / Fal Price'. Source: kie.aichecked |
| Platform fee | None; 'no hidden fees' and 'Failed Tasks $0' claimed | Source: kie.aichecked |
| Credits and free tier | $0.005 a credit (200 per dollar) on 582 of 590 rows; 80 free credits for new users; credits never expire | Source: kie.aichecked |
| Minimum top up and payment | Not shown on public pages; the billing page needs a login | Source: kie.aichecked |
| Models | 590 price rows across 22 maker labels: video, image, music (Suno) and chat models | Source: kie.aichecked |
| API formats | Task API with callbacks for media; Anthropic Messages at api.kie.ai/anthropic; OpenAI Responses at api.kie.ai/openai/v1; per-model OpenAI Chat Completions for Gemini | Source: docs.kie.aichecked |
| MCP | No official MCP server found in its docs; third-party wrappers exist | Source: docs.kie.aichecked |
| Agent Skills | Two official skills, kie-models and kie-chat-agents: npx skills add https://kie.ai | Source: docs.kie.aichecked |
| Coding agents | Setup guides for Claude Code, Codex CLI and Grok Build | Source: docs.kie.aichecked |
| Data retention and rate limit | Generated media 14 days; logs 2 months; 20 new generation requests per 10 seconds | Source: docs.kie.aichecked |
| Status page | None found; the homepage claims both 99.9% uptime and 99% availability | Source: kie.aichecked |
| Refunds | "Supported if needed" by email; no conditions or time window stated | Source: kie.aichecked |
| Website visits, September 2026 (kie.ai) | 3.1 million (+67.3% on August); 2nd of 10 on AICPB's AI model routers list; 6th of 20 on AICPB's fastest-growing list | Source: mp.weixin.qq.comchecked |
Answer first, then why, an example, the limits and what to do next.
Kie.ai is a real business with public terms and paying users, but it does not claim to be an authorised reseller of the models it sells. Its About page names two LLCs at a Denver address under terms in force since August 1, 2025, and it does not explain how it gets access to Google, OpenAI, Anthropic or Suno models.
Why
Its terms say the trademarks it shows do not imply endorsement by their owners, and its footer names a third entity, 'kie.ai Inc.'.
Kie.ai's list prices are lower than the makers' on every model compared here. Veo 3.1 Quality at 1080p costs $1.275 against Google's $3.20, Nano Banana Pro $0.09 per 1K image against $0.134, and Claude Opus 5.5 $1.60 / $8.00 per million tokens against Anthropic's $4 / $20.
Limits
Its own savings column often measures against fal's prices, its Claude price is below Anthropic's lowest published rate with no explanation, and Replicate is cheaper on Seedance 2.0.
Public reports about Kie.ai's reliability are mostly negative. It scores 2.1 out of 5 from 16 Trustpilot reviews, with complaints about jobs stuck at 99% while credits are used and hours of 500 errors; it has no public status page, and Kie.ai says credits for failed tasks are returned automatically.
Limits
Sixteen unverified reviews are a small sample, and no benchmark such as Artificial Analysis tracks Kie.ai.
Kie.ai does not publish an official MCP server in its docs. It offers two Agent Skills instead, kie-models and kie-chat-agents, installed with npx skills add https://kie.ai, plus setup guides for Claude Code, Codex CLI and Grok Build; third-party MCP wrappers exist but are not Kie.ai's own.
Next step
Check any community MCP package's code before giving it your API key.
Refunds from Kie.ai are possible but not guaranteed. Its refund policy says refunds are "supported if needed" and asks you to email support with your order number, but it gives no conditions or time window.
Kie.ai keeps generated media for 14 days and logs for two months, but its privacy policy does not say what happens to prompts and uploads. The policy covers only your email address, and the terms license your content to Kie.ai for operating and improving the service.
The main alternatives to Kie.ai are fal, Replicate and WaveSpeedAI for media, OpenRouter for language models, and the makers' own APIs. fal, Replicate and WaveSpeedAI charge $1.20 for an 8-second Veo 3.1 Fast clip against Kie.ai's $0.30, but they name their companies, publish status pages and appear in independent speed benchmarks.
This page is built from Kie.ai's own site, pricing table, docs and policies, the price pages of Google and Anthropic, the model pages of fal, Replicate and WaveSpeedAI, public reviews by AI Softing, Bitdoze and AI Insights News, and Trustpilot reviews, all checked on October 21, 2026. Kie.ai's claims are labelled as its own.
According to AICPB's traffic rankings, kie.ai had 3.1 million visits in September 2026, up 67.3% on August: 2nd among AI model routers and 6th on the fastest-growing list.
Kie.ai sells one API key and one credit balance for other companies' models. Its pricing table has 590 price rows across 22 maker labels: video (Veo 3.1, Kling 3.0, Seedance 2, Runway), images (Nano Banana Pro, GPT Image 2, Flux), Suno for music, and language models including Claude Opus 5.5, GPT-6.1 Sol and Gemini 4 Argon. Credits cost $0.005 each, 200 per dollar, on 582 of the 590 rows. New accounts get 80 free credits, about $0.40, and credits never expire. Kie.ai says failed tasks cost nothing.
Image, video and music jobs run as tasks: you get a task ID, then poll or receive a callback. The default limit is 20 new requests every 10 seconds; extra requests get a 429 error rather than a place in a queue.
The site is run by NEXUSAI SERVICES LLC and INNOLEAP AI LLC at a Denver, Colorado address, under terms effective August 1, 2025 and governed by Colorado law. Its footer reads "© 2026 kie.ai Inc.", a third name the About page does not explain.
Kie.ai says many models cost about 30% less than official prices, some far less. On the rows below it is lower than the makers and than the other three media services in most cases. Prices in US dollars, from each service's own pricing or model page, Google's Gemini API pricing (updated October 19, 2026) and Anthropic's pricing:
| Model and output | Maker's list | Kie.ai | fal | Replicate | WaveSpeedAI |
|---|---|---|---|---|---|
| Veo 3.1 Fast, 8 s at 720p with audio | 0.80 (Google) | 0.30 | 1.20 | 1.20 | 1.20 |
| Veo 3.1 Quality, 8 s at 1080p with audio | 3.20 (Google) | 1.275 | 3.20 | 3.20 | 3.20 |
| Nano Banana Pro, one 1K or 2K image | 0.134 (Google) | 0.09 | 0.15 | 0.15 | 0.14 |
| Nano Banana Pro, one 4K image | 0.24 (Google) | 0.12 | 0.30 | 0.30 | 0.24 |
| Kling 3.0 Standard, per second without / with audio | not confirmed | 0.07 / 0.10 (720p) | 0.084 / 0.126 | 0.168 / 0.252 | 0.084 / 0.126 |
| Seedance 2.0, per second at 720p | not confirmed | 0.205 | 0.3034 (with audio) | 0.18 | 0.24 |
| Claude Opus 5.5, per million input / output tokens | 4 / 20 (Anthropic) | 1.60 / 8.00 | not sold | not checked | not checked |
Against the makers that is 62.5% off Veo 3.1 Fast, 60% off Veo 3.1 Quality, 33% off Nano Banana Pro at 1K or 2K (50% at 4K) and 60% off Claude Opus 5.5. Claude Sonnet 5.5 is $0.80 / $4 against Anthropic's $2 / $10.
Four cautions. Kie.ai's Veo 3.1 Fast price is per clip of 4, 6 or 8 seconds, and its Veo page calls audio experimental. Its Claude Opus 5.5 price is below Anthropic's own Batch rate of $2 / $10, with no explanation. Its discount column is headed "Official / Fal Price" and often uses fal's price: for Veo 3.1 Fast it shows $1.20, so its claimed 75% saving is 62.5% against Google. And it is not cheapest everywhere: Replicate charges $0.18 a second for Seedance 2.0 at 720p against Kie.ai's $0.205. Kling's and BytePlus's own prices could not be confirmed.
Kie.ai's homepage, About page and terms do not say it is an authorised reseller or partner of Google, OpenAI, Anthropic, Suno or any other maker, and they do not say how it gets access. Its terms say the trademarks shown "do not imply any endorsement". The homepage mentions "smart fallback routing" without saying whether a request can be served by a different provider or model version. No public comparison of its outputs with the makers' own APIs was found.
Artificial Analysis does not track Kie.ai, and no independent speed or failure-rate measurement was found. The public reviews are light on numbers:
Trustpilot shows 2.1 out of 5 from 16 reviews, 81% one star. Reviewers report jobs stuck at 99% while credits were used (June 24 and October 2, 2026; the October reviewer puts it at about one generation in five), paid access without top models (August 2026), 8 to 10 hours of HTTP 500 errors after a large top-up (May 10, 2026), an image-to-image API that ignored reference images (May 15, 2026), and Sora 2 outages with support only in Asian hours (January 2026). Two five-star reviews (March 2026) praise low prices, non-expiring credits, the docs and a patient support reply. Kie.ai replied in July 2026 that it does not alter results and that credits for failed tasks are returned automatically; it replies to 7% of negative reviews. Hacker News has no discussion of it.
No official MCP server was found in Kie.ai's docs: its llms.txt index and AI agent section list none. Third-party wrappers exist, such as a community package listed on Glama and a guide from AnythingMCP.
Kie.ai does publish two Agent Skills, installed with npx skills add https://kie.ai and a KIE_API_KEY. kie-models lets a coding agent pick one of 200+ image, video, music and speech models, read its parameters, upload files, submit tasks and download results. kie-chat-agents configures coding agents to run on Kie.ai's chat models. The docs name Claude Code, Codex CLI and Grok Build.
For Claude Code, Kie.ai serves Anthropic Messages at https://api.kie.ai/anthropic: set ANTHROPIC_BASE_URL to it and ANTHROPIC_AUTH_TOKEN to your key. For Codex CLI it serves the OpenAI Responses API at https://api.kie.ai/openai/v1. Gemini models get per-model OpenAI-style Chat Completions endpoints; the Gemini 4 Argon one documents tool calling and a Google Search tool, and returns errors with HTTP 200 and a code in the body. The Claude Code guide does not say whether prompt caching, extended thinking or web search work, and no official SDKs are listed.
fal, Replicate and WaveSpeedAI charge more for Google's models but name their companies clearly, publish status pages and appear in Artificial Analysis' benchmarks; Replicate is also cheaper than Kie.ai on Seedance 2.0. For Claude and other chat models, OpenRouter bills at the providers' list prices plus a 5.5% fee on credit purchases. The makers' own APIs cost more but come with their data terms.
Anyone sending confidential data, anyone who needs an agreement with the model maker, and production apps without a fallback provider. For cheap experiments with image, video or music models, the main risk is the credits you put in.
Sources
Changelog
Kie.ai is a pay-per-call API for developers who want Veo 3.1, Kling, Seedance and other models behind one key, with credits that never expire. Higgsfield is a subscription web studio for creators and teams, with a separate API that undercuts Kie.ai on some models, such as Kling 3.0.
APIMODELS is a pay-as-you-go AI API relay that sells about 150 models through one key at its own prices, mostly below the makers' list prices, and spends the margin on backup channels: 98.62% of requests succeeded over 30 days, moderation rejections excluded. APIMODELS is our own product: it is made by the team behind GenGrades.
WaveSpeedAI is an API platform, run by a Singapore company and a Hong Kong company named in its terms, that sells 1,000+ image, video, audio and 3D models plus an LLM service on one balance. It matches fal on Kling 3.0, charges above Google's list for Veo 3.1 Fast and accepts Alipay and WeChat Pay.
Vercel AI Gateway is Vercel's hosted API for about 480 text, image, video and speech models behind one key, billed from prepaid credits at the providers' list prices with no markup, including when you bring your own keys. It became generally available on August 21, 2025, and gives each Vercel team 5 USD of free credit a month.
Replicate is an API for running AI models, owned by Cloudflare since a $57.4 million acquisition closed on December 1, 2025. It sells 100+ official models such as Veo 3.1, Nano Banana Pro and Kling 3.0 per output, and 50,000+ community models by the second of GPU time.
fal is a San Francisco API platform, run by Features & Labels, Inc., that sells per-output access to more than 1,000 image, video, audio and 3D models on one prepaid balance. Its open models are cheap, but Google partner models such as Veo 3.1 Fast and Nano Banana Pro cost more on fal than on Google's own API.