Kimi Moonshot API Access Outside China: Direct vs. Aggregator Routes (As of Aug 2026)
International developers can legitimately call Kimi models in August 2026 through two paths: Moonshot AI's official international platform platform.kimi.ai, or third-party aggregators like HeFu that resell Kimi K2.5, K2.6, and K3 through OpenAI-compatible endpoints. The official route works but carries phone-verification, billing, and compliance friction for non-China teams, which is why aggregator access has become the practical default for most overseas developers.
Official Access Status as of Aug 2026
Moonshot AI runs two fully separate platforms: platform.kimi.ai for international users and platform.moonshot.cn for mainland China. Accounts, billing profiles, and API keys are not interchangeable across the two; mixing them returns 401 invalid_authentication_error (Kimi official help center, retrieved Aug 2026). The international platform is operated by the Singapore legal entity Moonshot AI Pte. Ltd., which means overseas billing and support flow through a non-China parent (Moonshot terms of service, retrieved Aug 2026).
As of Aug 2026, self-service sign-up on platform.kimi.ai is open to developers with a non-China phone number and an international credit card. Higher rate limits and enterprise service-level agreements require a direct conversation with Moonshot's sales team; the eligibility and required documents are described on Moonshot's enterprise page (retrieved Aug 2026).
Geographic and Legal Restrictions on Direct Usage
Non-China developers who take the direct route typically hit three barriers:
- Phone verification. platform.moonshot.cn requires a mainland China mobile number that most overseas developers do not own. In practice, this locks out the China platform entirely (account registration requirements, retrieved Aug 2026).
-
Hardcoded endpoints. Community SDK providers have hardcoded the official provider to point at
api.moonshot.ai(the international URL). GitHub Issue #2883 documents that China-region API keys cannot be switched in via environment variables, producing repeated 401 errors even when the key is valid (ofox.ai blog, retrieved Aug 2026). - Payment and entity checks. International accounts pay in USD, but Moonshot excludes certain jurisdictions, and enterprise contracts require business registration documents before credit terms are granted; current currency and compliance terms are listed on Moonshot's billing documentation (retrieved Aug 2026).
These restrictions do not make direct access impossible — they make it slow and bureaucratic, which is the primary reason the aggregator market grew around the Kimi API in 2026.
Access via Third-Party API Aggregators
Aggregators give overseas developers a compliant shortcut: one API key, one OpenAI-compatible base URL, and credit-card payment in USD — no China phone number, no sales cycle. HeFu is one such aggregator and lists Kimi K2.5, K2.6, and K3 alongside OpenAI GPT-5.6 series, Claude Opus 5, DeepSeek-V4-Pro, and Gemini 3.6 Flash (HeFu models page, retrieved Aug 2026). Our HeFu AI API aggregation guide explains the unified key architecture, and our Hong Kong node overview covers routing for Asia-Pacific developers. For a side-by-side look at another flagship model through the same route, see our Claude Opus 5 API access article.
Market data supports the aggregator credibility: Kimi K2.6, released with open weights in Apr 2026, is already served by at least nine third-party API platforms, including OpenRouter, DeepInfra, and Fireworks, per DeepInfra's blog and Puter.js, both retrieved Aug 2026.
Pricing and Billing for Non-China Users
The official list prices as of Aug 2026 are:
- Kimi K2.5: $0.60 per million input tokens, $3.00 per million output tokens; cache-hit input at $0.10.
- Kimi K2.6: $0.95 / $4.00 per million tokens; cache-hit input at $0.16.
- Kimi K3: $3.00 / $15.00 per million tokens, released Jul 16, 2026, with a 1M-token context window.
Sources: Puter.js, BenchLM.ai (Jul–Aug 2026), and Moonshot's official pricing page (retrieved Aug 2026). The official Batch API charges 60% of the standard price for K2.7 Code, K2.6, and K2.5; K3 is not yet listed on the official Batch pricing page as of Aug 2026 (Moonshot batch pricing, retrieved Aug 2026).
Aggregators price differently. OpenRouter listed approximately $0.7448 input / $4.655 output for K2.6 as of Aug 2026, as shown on the OpenRouter model page. Fireworks matches the official $0.95 / $4.00; Requesty has promoted kimi-k3 at a 10% discount, i.e., $2.70 / $13.50 (Requesty pricing announcement, Aug 2026). HeFu uses a per-model rate card that adjusts with upstream costs, so the company does not publish fixed "forever" prices — check HeFu's official pricing page for current per-model rates and package terms.
Latency, Performance, and Routing Considerations
Direct calls from outside China are terminated at Moonshot's edge nodes in Singapore or the US West region; typical time-to-first-byte for short prompts is 100–300 ms (measured from Moonshot's status page, retrieved Aug 2026). The problem for developers relying on direct access is that these metrics can vary by region, and without local presence, performance may degrade. Aggregators often provide more consistent routing through optimized paths.
