Chinese LLM API Pricing Comparison (2026)

Across the major Chinese LLM families in 2026, DeepSeek V4-Flash anchors the floor at $0.14 input and $0.28 output per million tokens, while Qwen, GLM, Kimi, and MiniMax span roughly $0.06 to $3.90 depending on tier and the endpoint you route through. That last clause is the catch most pricing tables hide. The number you pay depends heavily on whether you hit a vendor's official international endpoint or a third-party router, and for everything except DeepSeek below, the prices we measured came through OpenRouter rather than a native key.

So treat this as two layers. DeepSeek figures are from its official endpoint and are solid. The rest are routed prices that often undercut the official international rate, which is useful to know but carries a needs_native_reverify flag until we confirm them on each vendor's own key.

Chinese LLM API pricing compared (verified 2026-06)

Here is the master table, per 1M tokens, USD. The source column is the part that matters: read it before you read the price.

OpenRouter is a routing layer that sells access to many models through one API and one bill. The headline takeaway: the cheapest credible general tier across the whole set is DeepSeek V4-Flash on its own endpoint, while the absolute floor on input belongs to GLM-4.7-Flash at $0.06 and Qwen3.5-Flash at $0.065 through routing.

Why routed prices can beat the official endpoint

The surprising part for most buyers is that a reseller can be cheaper than the vendor's own door. According to Alibaba Cloud, the Qwen international Model Studio rate for the flagship Max tier sits near $2.40 input and $12.00 output per million, yet the same family routed through OpenRouter measured $0.78 and $3.90. That is not a rounding gap. It is a structural one.

Two things drive it. First, the international endpoints vendors expose outside mainland China often price in a premium over the domestic Chinese rate, and a router that sources capacity closer to the cheaper origin passes part of that saving through. Second, routers run competitive marketplaces where several hosts bid to serve the same model, which compresses margin. The practical upshot: if you only ever checked the official global pricing page, you would overestimate Qwen and GLM costs by a wide margin.

The caveat cuts both ways. According to z.ai, GLM's own platform lists GLM-4.6 near $0.60 input and $2.20 output and offers a genuine free tier on its Flash models, which a router cannot match on the free side. So routing wins on paid throughput for some families and loses on free-tier access for others. This is exactly why every non-DeepSeek row above stays flagged for native re-verification.

What we measured first-hand (via OpenRouter)

Documentation gives you rate cards. We wanted real latency and real billed cost, so we ran small general and coding prompts against each family and recorded what came back. DeepSeek calls went to its official endpoint; every other family below was measured via OpenRouter, and those numbers carry the re-verify flag.

What surprised us was speed, not price. DeepSeek V4-Flash on the official endpoint came back in 0.7 to 2.3 seconds and was consistently the fastest model in the set. The routed families clustered slower: Qwen in the 2 to 3.6 second range, GLM and Kimi often past 7 seconds, and MiniMax around 5. We re-ran the slow ones to be sure, and the pattern held.

One trap worth naming. "Flash" in a model name does not mean fast. When we called GLM-4.7-Flash it ran a full 26.2 seconds and emitted 200 completion tokens, and Qwen3.5-Flash took 17.85 seconds while generating 2,831 completion tokens on a two-sentence prompt. The reasoning and thinking tiers behave the same way: Kimi K2-Thinking spent 18.51 seconds producing 553 tokens. That blown-up output is the hidden cost, because a low per-token rate stops mattering when the model writes ten times more tokens than you expected.

A cross-family pattern fell out of this. DeepSeek V4-Flash was simultaneously the cheapest credible general tier and the fastest model we measured, a both-wins result that is rare in this category. Most families force a trade between the two.

How to read this table without getting burned

Pick by workload, not by the headline number. For high-volume chat where latency and floor price both matter, DeepSeek V4-Flash on the official endpoint is the safe default. For the absolute cheapest input tokens you can find a real free allotment, GLM's Flash tier on z.ai is worth testing. For long-context document work, Qwen and MiniMax have larger context stories that this short test did not stress.

Whatever you choose, confirm the price on the endpoint you will actually use in production. According to DeepSeek API Docs, DeepSeek publishes its rates directly and we verified them on the official endpoint, but the routed prices for the other four families came through a marketplace and can move when hosts change. For deeper per-family breakdowns, our DeepSeek API pricing hub and the Qwen API pricing guide carry the tier-by-tier detail.

The honest summary: there is no single cheapest Chinese LLM API, only a cheapest one for your specific call shape and the endpoint you route it through.

FAQ

What is the cheapest Chinese LLM API in 2026? On its official endpoint, DeepSeek V4-Flash at $0.14 input and $0.28 output is the cheapest credible general tier and the fastest we measured. Through OpenRouter routing, GLM-4.7-Flash ($0.06 input) and Qwen3.5-Flash ($0.065) go lower, but those are routed figures pending native re-verification.

Why are OpenRouter prices lower than the official vendor pages? Vendors' international endpoints often carry a premium over domestic Chinese rates, and routers run competitive marketplaces where multiple hosts bid to serve the same model. Both compress the routed price below the official global rate for families like Qwen and GLM.

Are these prices from the official Chinese vendor APIs? Only the DeepSeek rows are. DeepSeek was measured on its official api.deepseek.com endpoint. Qwen, GLM, Kimi, and MiniMax prices and latencies were measured via OpenRouter and are flagged needs_native_reverify until confirmed on each vendor's own key.

Does a "Flash" model mean it is fast? No. In our tests GLM-4.7-Flash took 26.2 seconds and Qwen3.5-Flash took 17.85 seconds while emitting 2,831 tokens. The naming refers to tier positioning, not measured latency, and reasoning tiers can run even longer.

Which Chinese LLM is fastest? DeepSeek V4-Flash on its official endpoint, at 0.7 to 2.3 seconds across our calls. Routed families ran slower: Qwen 2 to 3.6 seconds, MiniMax around 5, GLM and Kimi often past 7 seconds.

---

This is part of the best Chinese LLM API hub, which ranks the full field.

Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing. Last verified: 2026-06-26.

<!-- METADATA { "title": "Chinese LLM API Pricing Compared in 2026", "slug": "chinese-llm-api-pricing-comparison", "meta_description": "Master 2026 pricing table for DeepSeek, Qwen, GLM, Kimi, MiniMax APIs per 1M tokens. DeepSeek official; others OpenRouter-routed. Why routed beats official.", "focus_keyword": "chinese llm api pricing comparison", "secondary_keywords": ["chinese llm api pricing", "deepseek vs qwen vs glm pricing", "cheapest chinese llm api", "openrouter vs official pricing", "chinese ai model api cost"], "tags": ["DeepSeek", "Qwen", "GLM", "Kimi", "MiniMax", "API Pricing"], "category": "Pricing", "cluster_id": "best-chinese-llm-api", "cluster_role": "micro", "hub_slug": "best-chinese-llm-api", "evidence_file": "clients/china-llm-aggregator/articles/best-chinese-llm-api-evidence.json", "needs_native_reverify": true, "verified_until": "2026-09-24", "author_name": "Kevin Fan", "author_title": "Customer Success Manager", "author_linkedin": "", "author_expertise": ["Chinese LLM ecosystem", "AI infrastructure pricing", "model benchmarking", "cross-border AI compliance"], "faq_pairs": [ {"q": "What is the cheapest Chinese LLM API in 2026?", "a": "On its official endpoint, DeepSeek V4-Flash at $0.14 input and $0.28 output is the cheapest credible general tier and the fastest we measured. Through OpenRouter routing, GLM-4.7-Flash ($0.06 input) and Qwen3.5-Flash ($0.065) go lower, but those are routed figures pending native re-verification."}, {"q": "Why are OpenRouter prices lower than the official vendor pages?", "a": "Vendors' international endpoints often carry a premium over domestic Chinese rates, and routers run competitive marketplaces where multiple hosts bid to serve the same model. Both compress the routed price below the official global rate for families like Qwen and GLM."}, {"q": "Are these prices from the official Chinese vendor APIs?", "a": "Only the DeepSeek rows are. DeepSeek was measured on its official api.deepseek.com endpoint. Qwen, GLM, Kimi, and MiniMax prices and latencies were measured via OpenRouter and are flagged needs_native_reverify until confirmed on each vendor's own key."}, {"q": "Does a Flash model mean it is fast?", "a": "No. In our tests GLM-4.7-Flash took 26.2 seconds and Qwen3.5-Flash took 17.85 seconds while emitting 2,831 tokens. The naming refers to tier positioning, not measured latency, and reasoning tiers can run even longer."}, {"q": "Which Chinese LLM is fastest?", "a": "DeepSeek V4-Flash on its official endpoint, at 0.7 to 2.3 seconds across our calls. Routed families ran slower: Qwen 2 to 3.6 seconds, MiniMax around 5, GLM and Kimi often past 7 seconds."} ], "external_links_used": [ {"url": "https://api-docs.deepseek.com/quick_start/pricing/", "source_name": "DeepSeek API Docs – Pricing", "claim": "DeepSeek publishes official V4-Flash rates $0.14/$0.28 per 1M; verified on official endpoint"}, {"url": "https://www.alibabacloud.com/help/en/model-studio/models", "source_name": "Alibaba Cloud Model Studio", "claim": "Qwen international Model Studio flagship Max tier near $2.40 input / $12.00 output per 1M"}, {"url": "https://z.ai/", "source_name": "z.ai", "claim": "GLM-4.6 official rate near $0.60/$2.20 per 1M; genuine free tier on Flash models"} ], "internal_links_used": [ {"url": "/blog/best-chinese-llm-api/", "anchor_text": "best Chinese LLM API hub", "type": "hub"}, {"url": "/blog/deepseek-api-pricing/", "anchor_text": "DeepSeek API pricing hub", "type": "cluster-hub"}, {"url": "/blog/qwen-api-pricing/", "anchor_text": "Qwen API pricing guide", "type": "cluster-hub"} ], "first_hand_evidence": { "source": "best-chinese-llm-api-evidence.json runs (qwen3-max_general, qwen-plus_general, glm-4.6_general, kimi-k2_general, minimax-m2_general; latency leaderboard)", "measured": "Routed billed costs/latency: Qwen3-Max $0.00027534/3.22s; Qwen-Plus $0.00008320/2.47s; GLM-4.6 $0.00036376/8.58s; Kimi K2 $0.00015852/7.03s; MiniMax-M2 $0.00020995/5.08s. DeepSeek V4-Flash official 0.7-2.3s fastest. GLM-4.7-Flash 26.24s, Qwen3.5-Flash 17.85s/2831 tok, Kimi K2-Thinking 18.51s/553 tok.", "disclosure": "DeepSeek via official api.deepseek.com; all other families via OpenRouter (needs_native_reverify)", "captured": "2026-06-26" }, "images_status": "spec-only (not generated; FAL_API_KEY unset)", "images": [ {"position": "featured", "type": "generated", "prompt": "Clean editorial comparison chart of five Chinese LLM families (DeepSeek, Qwen, GLM, Kimi, MiniMax) plotting input vs output price per 1M tokens, with DeepSeek official endpoint highlighted and the other four labeled OpenRouter-routed. Blue and amber palette, minimal background, 16:9.", "alt": "Comparison chart of DeepSeek, Qwen, GLM, Kimi and MiniMax API input and output prices per 1M tokens, DeepSeek measured on its official endpoint and the others measured via OpenRouter"} ] } -->