YKH.AI Review: Pricing & Comparison

Pricing reduced to its simplest form

YKH.AI’s (ykh.ai) product philosophy can be summed up in one line: strip away everything except price and performance. There’s no flashy marketing page on the homepage, no sprawling feature-comparison matrix, and even the plan names are kept to a bare minimum: Lite and Pure Pro.

Two tiers, no ambiguity:

• Lite: ¥0.25/million tokens
• Pure Pro: ¥0.5/million tokens

There’s no tiered pricing table, no per-model price list, and no minimum top-up requirement. Choosing between Lite and Pro comes down to one question: how demanding is the task? Routine work — summarization, format conversion, batch classification — fits Lite. Anything that needs stronger reasoning, longer context, or more complex instruction-following goes on Pro. This “two tiers cover everything” design removes the “which plan do I actually need” decision fatigue that’s common at most relay platforms — plenty of comparable services list a dozen-plus models each individually priced, which arguably adds more choice-overload than it solves. YKH.AI went the opposite direction.

Where ¥0.25/M sits on the market map

To understand how aggressive YKH.AI’s pricing really is, it helps to put it on a comparison grid:

Platform/tier	Reference price (per million tokens, approx.)	Positioning
YKH.AI Lite	¥0.25	Rock-bottom low price
YKH.AI Pure Pro	¥0.5	Low price + higher performance
SBGPT	Converted at ~0.4–0.6¥/USD	Budget tier
ZHTec API	Converted at ~0.5–0.6¥/USD, VIP tiering	Budget tier, with VIP layers
Claude Sonnet 4.6 official price	~¥21.6 (input) / ¥108 (output)	Official pricing baseline

Even just looking at input pricing, YKH.AI’s Lite tier is roughly one-hundredth of Claude’s official price — that’s not a “discount,” it’s a fundamentally different pricing logic. Low prices at relay platforms generally come from one of a few sources: bulk-purchase discounts from upstream providers, minimal operating overhead from a stripped-down team, or (for some platforms) access via unofficial/reverse-engineered channels that undercut official cost structures. YKH.AI’s site doesn’t disclose the source of its pricing structure, and whether the Lite tier’s ultra-low price involves any unofficial channel is something users need to weigh and accept the risk of themselves — a reasonable default posture for any low-price relay, not a specific accusation against YKH.AI.

Reading the relay industry’s pricing logic through YKH.AI

Understanding YKH.AI’s price also means understanding the pricing vocabulary common across the whole relay industry. Providers often describe their price relative to official pricing using a “multiplier”: a multiplier of 1 means at-parity with official pricing, 0.8 means a 20% discount, and anything above 1 means a markup. Based on industry observation, relays that buy in bulk through official channels and resell typically land in the 0.8–1.5 multiplier range, while platforms tapping unofficial or reverse-engineered channels can go as low as 0.05–0.3 — that is, 5% to 30% of official price.

Converting YKH.AI Lite’s ¥0.25/M price into multiplier terms clearly lands it in the extreme-low-multiplier bracket. That doesn’t necessarily mean it’s using unofficial channels, but this price range is generally associated in the industry with “channel cost squeezed to the bare minimum” — worth knowing before you try to understand why the price can be this low. Chinese financial and tech media have repeatedly warned about the relay industry as a whole in recent years: the core message is that low prices often come with a trade-off, and buyers shouldn’t just look at the number without understanding the channel logic and risk behind it. Staying level-headed rather than impulsively loading up a large balance is worth more than the discount itself.

Reading the operator from the product design

Minimalism is itself a signal. A product with only two price tiers, no complex back-end features, and no large-scale marketing push usually corresponds to one of a few operating realities: a small team or solo operator; a volume-over-service business model; or most of the resources going into cost control rather than user-experience polish.

None of those are inherently bad — plenty of excellent developer tools started as “one person’s minimalist product.” But for users, it’s worth setting expectations accordingly: feature iteration is likely slower without a dedicated product team; edge cases (billing disputes, model anomalies) may not be handled with the rigor of a platform with a full-time support staff; and the long-term roadmap generally isn’t public, making it hard to tell whether this is a “stable small-and-focused operation” or a “short-term arbitrage play.”

The most reliable way to judge whether a platform with limited transparency like YKH.AI is trustworthy isn’t reading the marketing copy — it’s sustained small-scale observation: use it for a few weeks and watch whether pricing stays stable, whether responses stay consistent, and whether support responds when something goes wrong. That tells you more than any single review ever could.

Where does the low price come from — a question worth asking

For budget-conscious developers, ¥0.25/M is undeniably tempting, but it’s worth thinking through three questions before committing.

First, how long can this price last? Low-price relays’ lifespans are often tied to subsidies and channel costs — if upstream costs rise or channel access tightens, prices can jump sharply, or the platform can simply shut down. China’s relay-station space has already seen multiple cases of operators disappearing or overnight price hikes. This isn’t an accusation against YKH.AI specifically — it’s a structural risk of the entire low-price relay tier, and anyone choosing this price bracket should go in with that expectation.

Second, who do you turn to when something goes wrong? A minimalist design implies limited support resources. If an account gets suspended, a request behaves oddly, or a balance disappears, whether there’s an adequate escalation path and response time is a real question mark for low-price platforms generally, and YKH.AI’s site doesn’t publish detailed support-channel information either.

Third, is the public information sufficient to warrant trust? While preparing this review we specifically searched for third-party reviews, community discussion, and user feedback about YKH.AI. As of July 2026, public search results were very limited — we found no independent in-depth review or evidence of a sizable user community. That doesn’t mean YKH.AI is unreliable, but it does mean “test small before you commit, then keep watching” matters more here than on platforms with more public transparency.

What “performance-first” actually means

YKH.AI’s marketing copy includes the phrase “performance-first,” but the site doesn’t publish specific benchmark numbers or latency test reports. Based on the Lite/Pure Pro tier design, “performance-first” more likely refers to a trade-off in model routing strategy: the Lite tier routes to lower-cost, faster-responding but more basic models or channels, while Pure Pro routes to stronger but roughly twice-as-expensive channels — letting users choose their tier to balance “speed vs. cost” and “capability vs. cost,” rather than a claim that the platform runs faster overall than comparable services.

This reading is a reasonable inference from the product design, not an official statement. In practice, we’d recommend running the same prompt on both Lite and Pure Pro yourself and comparing output quality and response time before deciding which tier to use for which task type. This “test before trusting the marketing copy” mindset applies to nearly every relay with limited public documentation — YKH.AI is just a fairly typical example of the category.

Who it fits, who it doesn’t

Good fit:

• Individual developers’ day-to-day dev/test work — high call volume, low per-task stakes, price is the dominant variable
• Batch data processing, scraping/cleanup, format conversion — high error tolerance, occasional failures are manageable
• Teams that already have a primary platform and just want a cheaper “shadow backup” for staged testing
• Personal projects that don’t need enterprise features (audit logs, multi-member collaboration, invoicing)

Poor fit:

• Enterprise production environments — no public SLA commitment, no invoice/compliance support
• Team collaboration scenarios needing multi-member management or per-project usage auditing
• Core business paths where stability matters more than price
• Scenarios that need timely human support when something goes wrong

The dividing line is pretty clear: YKH.AI suits “it’s fine if this breaks” edge-case work, not “if this goes down, delivery is directly impacted” core business paths.

Cost estimates for three real-world scenarios

Putting the abstract pricing into concrete scenarios makes it easier to judge whether YKH.AI fits your use case:

Scenario 1: Batch content tagging. Say you need sentiment classification and keyword extraction across 100,000 user reviews, averaging 200 input tokens and 50 output tokens each. On the Lite tier, total cost works out to roughly 100,000 × 250 tokens ÷ 1,000,000 × ¥0.25 — a handful of yuan. At this scale, running the same job through official Claude or GPT pricing could cost one to two orders of magnitude more. This is exactly where low-price platforms like YKH.AI earn their keep.

Scenario 2: Personal project prototyping. You’re building a weekend vibe-coding project and need to call the model frequently to iterate on prompts — potentially thousands of trial-and-error calls a day. Price-sensitive, low bar for per-call correctness — YKH.AI Lite’s per-attempt cost is nearly negligible, well suited to this “iterate fast, fail cheap” development rhythm.

Scenario 3: Fallback backup for a primary platform. If your primary platform is a pricier, more stable relay, configuring YKH.AI as a “degraded fallback” — auto-switching over when the primary is rate-limited or lagging — trades cost savings for a bit of resilience. This is a reasonable pattern, but you need to build proper error handling and failover logic into your code, since you can’t assume the two platforms’ response formats and error codes match exactly.

Why not just use the official API directly

For developers in mainland China, registering directly with Anthropic or OpenAI has its own barriers: you typically need an overseas or dual-currency credit card to bind payment, official accounts commonly enforce regional restrictions and risk-control rules that put off individual developers, and if your account gets suspended or a payment fails, official support response times and English-only communication aren’t especially friendly. This is precisely why domestic relays like YKH.AI exist — pay via Alipay/WeChat Pay, access via a mainland direct connection, and make “can I actually use the model” a non-issue.

But that convenience comes at a cost: official API pricing and service come with clear contractual terms; a relay is an extra layer of trust intermediation. Users are effectively trading “lower price” for “fewer guarantees.” YKH.AI pushes that trade-off to the price-sensitive extreme — the lowest price in exchange for a “good enough to work” experience, with no claim to or emphasis on stability or guarantees. Understanding this trade-off tells you more about whether YKH.AI fits your actual needs than the price number alone.

FAQ

Which coding tools can connect to it? The site’s documentation centers on OpenAI-compatible format, so in theory any AI coding tool that supports a custom base_url (Claude Code, Cursor, Cline, etc.) can connect with a config change — but actual compatibility and parameter support need real-world testing, since there’s no per-tool integration guide. If a tool-specific parameter (extended context, tool-calling format) throws an error, you’ll need to figure out yourself whether it’s a tool compatibility issue or a platform limitation.

Is identity verification required? The site doesn’t mention a mandatory verification requirement; the actual signup flow is authoritative, and cross-border compliance/verification requirements can shift with policy.

How are billing and invoices handled? Given its minimalist positioning, YKH.AI most likely doesn’t offer corporate invoicing. If your reimbursement process requires a formal invoice, a platform like DuckCoding that explicitly supports invoicing is a better fit.

What happens if the platform suddenly shuts down? This is a risk shared by all small relays with limited public information. Don’t keep a large unused balance sitting in your account — top up in small amounts as needed to limit potential losses if the platform stops operating.

What payment methods are supported? Based on common practice among domestic relays, YKH.AI most likely supports at least Alipay or WeChat Pay — check the top-up page in your console after signing up for specifics. There’s no public mention of credit card or crypto support, so international users or cross-border teams should confirm before committing.

Is there a free trial credit? Nothing in the available information suggests YKH.AI offers a signup bonus, unlike SiliconFlow, OpenRouter, and others that provide free trial credit. If you want to test prompts at zero cost, validate your logic on a platform with free credit first, then migrate to YKH.AI for cost-efficient bulk execution — part of why this review keeps stressing “small top-up first, scale later.”

A one-line decision tree

If you’re still on the fence about YKH.AI, here’s a quick way to decide: need invoicing and a corporate procurement process → go with DuckCoding instead of YKH.AI; expect call volume to grow quickly with your business and need elastic plans → look at ZHTec or AiGoCode; just want to run a personal project or batch data job as cheaply as possible and can tolerate some uncertainty → YKH.AI is a candidate, but test with a small top-up first rather than committing a large sum upfront. There’s no universally correct answer here — the core logic is weighing “price priority” against “certainty priority” based on your project’s nature and risk tolerance.

How to connect

YKH.AI is OpenAI-compatible, so migration cost is close to zero:

from openai import OpenAI

client = OpenAI(
    api_key="your YKH.AI key",
    base_url="https://api.ykh.ai/v1"  # check the official site for the current endpoint
)

response = client.chat.completions.create(
    model="claude-fable-5-20260702",  # or another supported model ID
    messages=[{"role": "user", "content": "Batch task: extract keywords from the following text"}]
)

For projects already built on the OpenAI SDK, switching to YKH.AI typically only requires changing the base_url and api_key values — no code restructuring needed, which is exactly why OpenAI-compatible relays like this are called “painless migrations.”

A pre-flight checklist for low-price platforms

Given YKH.AI’s limited transparency, it’s worth running through this checklist before real usage:

1. Small top-up first: don’t commit a large sum upfront — test your workflow at small scale and observe response quality and stability
2. Watch for billing changes: low-price platforms can adjust pricing at any time — build a habit of periodically checking your bill and remaining credit
3. Have a Plan B: don’t single-source core business on any one low-price relay, especially one with limited public information
4. Keep request logs locally: if a dispute arises, having local request/response logs is the basis for any recourse
5. Verify the model ID matches the official version: low-price platforms sometimes pass off an older model as the latest — run a quick capability check to confirm the model is performing as expected

This checklist isn’t specific to YKH.AI — it’s a generally useful routine for any low-price relay with limited public information, worth treating as a default habit rather than a one-time setup task.

Where it sits among low-price competitors

YKH.AI isn’t alone in the ultra-low-price bracket. A few comparable competitors:

• SBGPT (0.4–0.6¥/USD, Azure-grouped pricing): prices via the official Azure channel grouping, in a similar range to YKH.AI, but generally seen as somewhat more credible given its Azure affiliation compared to a fully independent small operator
• ZHTec API (0.5–0.6¥/USD, VIP tiers): additional VIP-tier discounts, more flexible than YKH.AI’s flat two-tier structure for usage that scales up with your business
• AiGoCode (reverse-engineered low price + monthly plans): offers monthly plan options — for stable, predictable call volume, the monthly total may end up cheaper than YKH.AI’s pure pay-as-you-go
• DuckCoding ($1 signup credit, corporate invoicing supported): if you need an invoice for corporate reimbursement, DuckCoding is the more compliant choice — YKH.AI currently shows no sign of this feature

Dimension	YKH.AI	SBGPT	ZHTec	DuckCoding
Price range	Lowest (from ¥0.25)	Low	Low	Low-mid
Plan structure	Two tiers, simplest	Azure-grouped	VIP tiers	Tiered pricing
Corporate invoicing	Not seen	Not seen	Not seen	Supported
Best fit	Individual devs testing	Users comfortable in the Azure ecosystem	Users with growing call volume	Developers needing reimbursement

YKH.AI’s differentiator is its near-obsessive minimalism — two prices, one website, nothing extra getting in the way of the decision. If what you want is “cheap and it works,” with no add-ons required, YKH.AI’s product logic is the most direct among these alternatives. But if you value compliant invoicing, more flexible plan structures, or more established community reputation, the alternatives above may be a safer bet — worth testing each at small scale before committing seriously.

Bottom line

YKH.AI carves out a distinct spot in the low-price bracket with its radically simplified two-tier pricing (Lite ¥0.25/M, Pure Pro ¥0.5/M): cheap, direct, no runaround. Mainland direct connect and an OpenAI-compatible interface lower the integration bar, making it worth considering for budget-constrained, error-tolerant personal development use cases.

Its weaknesses are just as clear: little public information or third-party review coverage, no enterprise features or public SLA, and no observable track record to confirm the price will hold long-term. We’d position YKH.AI as a “low-cost testing ground” rather than a production-environment primary — test small, keep watching, and keep a backup plan, which is the sensible posture for any low-transparency low-price platform.

Within the broader AI API relay landscape, YKH.AI represents a typical product at the very bottom of the market’s price ladder: likely a small team, with operating focus entirely on driving price down rather than polishing brand or service experience. That’s a legitimate niche — there will always be users whose core need is “run the most tasks for the least money,” and YKH.AI serves that segment precisely. But if your project has reached the stage where you’re delivering to external stakeholders and need to be accountable for stability, shifting budget toward the other platforms covered in this review is usually worth the added certainty. The 3.8 rating reflects exactly this trade-off: price is near-perfect, but the lack of trust, transparency, and guarantees pulls down the composite score — a fair position for a price-driven rather than trust-driven product.

Information verified 2026-07-11. Pricing reflects the live ykh.ai site at time of writing; as of the verification date, no more detailed independent review or operating history was found through public channels — we recommend verifying independently before making any long-term commitment.

Related reviews

• Boxying: lightweight domestic relay, mainstream models, quick to get started
• Cooper-API: clean interface, friendly to individual developers, mainland direct connect to mainstream models
• DuckCoding: $1 signup credit, tiered pricing, corporate invoicing supported, focused on coding use cases
• UU API: MAX account pool, image generation at ¥0.04/image, ¥1 new-user bonus