Cheapest Chinese LLM API (2026)

The cheapest Chinese LLM API by sticker price is GLM-4.7-Flash at $0.06 per million input tokens (free on z.ai direct), edging Qwen3.5-Flash at $0.065, but the cheapest credible all-rounder is DeepSeek V4-Flash at $0.14, because per-token rates do not equal per-task cost. That last gap is the part most buyers miss, and our own test makes it concrete: the headline-cheapest tier produced the most expensive single call we measured.

If you are scanning for the lowest number, here is the ranked picture across every family we verified. One caution before the table: all prices except DeepSeek were measured via OpenRouter, so they reflect that routing layer rather than each vendor's native endpoint. Treat them as a directional ranking that still needs a native re-check.

Cheapest Chinese LLM API ranked (verified 2026-06)

The spread from top to bottom is roughly 4x on input, which sounds decisive until you look at output rates and actual token counts. GLM-4.7-Flash wins on input but charges $0.40 on output, more than DeepSeek V4-Flash and MiniMax-M2. So the "cheapest" label flips depending on whether your workload is input-heavy (long prompts, short answers) or output-heavy (short prompts, long generations). That is the first reason a ranked table alone misleads.

According to DeepSeek API Docs, DeepSeek V4-Flash bills input at $0.14 and output at $0.28 per million tokens on the official endpoint, and that is the one row here we measured natively rather than through a router. It sits fifth on raw input price, but it is the only sub-$0.30 tier that is both natively priced and, in our testing, the fastest to respond.

The cheapest per-token model is not the cheapest per task

Here is the trap. A "Flash" tier with the lowest sticker rate can still cost more on a real job if it generates far more output tokens to answer the same question. Cheap-per-token plus verbose equals expensive-per-task. The naming does not warn you: "Flash" implies fast and lean, and in two cases it was neither.

Per-task cost is the billed dollar amount for one complete request, not the per-million-token rate, and it folds in how many output tokens a model actually emits to finish the job. A model at half the per-token price that writes three times the tokens loses. This is the metric that decides your monthly bill, and it is the one the ranked table cannot show you.

First-hand evidence: what we measured (via OpenRouter)

We ran the same short general prompt across families and recorded billed cost and latency. Every non-DeepSeek number below was measured via OpenRouter on 2026-06-26 and is logged in our evidence pack; the DeepSeek anchor was a parallel call on its official endpoint. What surprised us was where the "cheapest" model landed.

Qwen3.5-Flash is second-cheapest per token at $0.065, yet on this prompt it ran away to 2,831 completion tokens and billed $0.00073866. That is the single most expensive general call in the set, roughly 20x the $0.00003672 that Qwen3-32B cost for a 120-token answer to a similar question. The cheap rate did not save money because the model would not stop talking, and at 17.85 seconds it was nowhere near "flash" on latency either.

The DeepSeek anchor tells the other half of the story. Across the cluster's runs, DeepSeek V4-Flash answered in 0.7 to 2.3 seconds, consistently the fastest family we tested, while staying at $0.14 input. Cheap and fast in the same model is rare; that combination is why it earns the all-rounder verdict over models that win on a single column.

Free tiers change the math

According to z.ai, GLM offers a genuine free tier on its Flash models served directly, which is why GLM-4.7-Flash can show as $0.06 through OpenRouter yet cost nothing on the vendor's own endpoint. For low-volume or prototyping work, a real free tier beats any paid sub-$0.10 rate. According to MiniMax Platform, MiniMax publishes its own native rate card that runs slightly above the OpenRouter figures we measured, so a native re-check is mandatory before you commit a budget. The practical upshot: the cheapest option for you depends on whether you can live inside a free tier's limits or need paid throughput.

Verdict

Choose GLM-4.7-Flash if your workload fits a free tier and your prompts are input-heavy. Choose Qwen3-32B if you want the cheapest paid call that stays terse. Avoid Qwen3.5-Flash for cost-sensitive jobs until you cap its output, because it overspends on verbosity. For most teams that want one model that is cheap, fast, and predictable, DeepSeek V4-Flash at $0.14 is the cheapest credible all-rounder. For the full cross-family breakdown, see the best Chinese LLM API hub, and for native rate cards check the DeepSeek pricing hub or the GLM pricing hub.

FAQ

What is the cheapest Chinese LLM API per token? GLM-4.7-Flash at $0.06 per million input tokens (free on z.ai direct), followed by Qwen3.5-Flash at $0.065 and Qwen3-32B at $0.08. These were measured via OpenRouter and need a native re-check; the cheapest per token is not always the cheapest per task.

Is DeepSeek V4-Flash the cheapest Chinese LLM API? Not on raw input price; it ranks fifth at $0.14. But it is the cheapest credible all-rounder because it was the fastest family we measured (0.7 to 2.3 seconds) and its output rate ($0.28) undercuts several models that look cheaper on input alone.

Why did Qwen3.5-Flash cost more despite a low per-token rate? On our test it generated 2,831 output tokens for a short prompt and billed $0.00073866, the most expensive general call in the set. A low per-token rate does not help when the model emits far more tokens to answer the same question.

Are these prices from official vendor endpoints? Only the DeepSeek figures are native (api.deepseek.com). Qwen, GLM, Kimi, and MiniMax prices were measured via OpenRouter, so they carry needs_native_reverify and may differ from each vendor's own rate card.

---

This is part of the best Chinese LLM API hub. For native rate cards see the DeepSeek pricing hub.

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

<!-- METADATA { "title": "The Cheapest Chinese LLM APIs, Ranked (2026)", "slug": "cheapest-chinese-llm-api", "meta_description": "GLM-4.7-Flash ($0.06) and Qwen3.5-Flash ($0.065) win on sticker price, but DeepSeek V4-Flash ($0.14) is the cheapest credible all-rounder. We tested per task.", "focus_keyword": "cheapest chinese llm api", "secondary_keywords": ["cheapest chinese llm", "cheapest llm api", "cheapest chinese ai api", "lowest cost chinese llm"], "tags": ["Chinese LLM", "API Pricing", "DeepSeek", "Qwen", "GLM", "MiniMax"], "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 per token?", "a": "GLM-4.7-Flash at $0.06 per million input tokens (free on z.ai direct), followed by Qwen3.5-Flash at $0.065 and Qwen3-32B at $0.08. These were measured via OpenRouter and need a native re-check; the cheapest per token is not always the cheapest per task."}, {"q": "Is DeepSeek V4-Flash the cheapest Chinese LLM API?", "a": "Not on raw input price; it ranks fifth at $0.14. But it is the cheapest credible all-rounder because it was the fastest family we measured (0.7 to 2.3 seconds) and its output rate ($0.28) undercuts several models that look cheaper on input alone."}, {"q": "Why did Qwen3.5-Flash cost more despite a low per-token rate?", "a": "On our test it generated 2,831 output tokens for a short prompt and billed $0.00073866, the most expensive general call in the set. A low per-token rate does not help when the model emits far more tokens to answer the same question."}, {"q": "Are these prices from official vendor endpoints?", "a": "Only the DeepSeek figures are native (api.deepseek.com). Qwen, GLM, Kimi, and MiniMax prices were measured via OpenRouter, so they carry needs_native_reverify and may differ from each vendor's own rate card."} ], "images_status": "spec-only (not generated; FAL_API_KEY unset)", "images": [ {"position": "featured", "type": "generated", "prompt": "Clean editorial ranked bar chart of cheapest Chinese LLM APIs by input price, GLM-4.7-Flash to MiniMax-M2, with one bar flagged for runaway per-task cost. Blue and amber palette, minimal background, 16:9.", "alt": "Ranked bar chart of the cheapest Chinese LLM APIs by input price from GLM-4.7-Flash to MiniMax-M2, with a flag marking the model whose per-task cost ran away despite a low per-token rate"} ], "external_links_used": [ {"url": "https://api-docs.deepseek.com/quick_start/pricing/", "source_name": "DeepSeek API Docs - Pricing", "claim": "DeepSeek V4-Flash native input $0.14 / output $0.28 per 1M tokens"}, {"url": "https://z.ai", "source_name": "z.ai", "claim": "GLM Flash models have a genuine free tier on the native endpoint"}, {"url": "https://platform.minimax.io", "source_name": "MiniMax Platform", "claim": "MiniMax native rate card runs slightly above OpenRouter-measured figures"} ], "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 pricing hub", "type": "cluster-hub"}, {"url": "/blog/glm-api-pricing/", "anchor_text": "GLM pricing hub", "type": "cluster-hub"} ], "first_hand_evidence": { "source": "best-chinese-llm-api-evidence.json runs: qwen3.5-flash_general, minimax-m2_general, qwen3-32b_general, qwen-plus_general, deepseek_v4flash_coding_anchor", "measured": "Qwen3.5-Flash 2831 out / $0.00073866 / 17.85s; MiniMax-M2 200 out / $0.00020995 / 5.08s; Qwen3-32B 120 out / $0.00003672 / 3.61s; Qwen-Plus 94 out / $0.00008320 / 2.47s; DeepSeek V4-Flash 86 out / 1.96s official endpoint. Non-DeepSeek measured via OpenRouter.", "captured": "2026-06-26" }, "entity_triples": [ {"subject": "GLM-4.7-Flash", "predicate": "has input price", "object": "$0.06 per 1M tokens (free on z.ai)"}, {"subject": "DeepSeek V4-Flash", "predicate": "is", "object": "cheapest credible all-rounder Chinese LLM API"} ] } -->