The Chinese AI Models That Cost 1/10th the Price

Most teams default to GPT-5, Claude Sonnet, or Gemini Pro. These charge $1.25–$5.00 per million input tokens and $10.00–$25.00 per million output tokens. Chinese AI labs have shipped models that match or beat those benchmark scores at a fraction of the price — some as low as $0.10/1M input and $0.30/1M output.

Nine Chinese labs — DeepSeek, Xiaomi, Z AI (Zhipu), Kimi (Moonshot AI), MiniMax, Alibaba (Qwen), Baidu (ERNIE), ByteDance Seed, and KwaiKAT — now have models in our database with published pricing and benchmark scores. Fifteen of those models score 39+ on the Intelligence Index, putting them in the same tier as Claude 4.5 Sonnet and GPT-5.

Where the prices come from

All prices in this post are list prices per million tokens — $/1M input and $/1M output — published by each vendor. This is the industry-standard pricing unit and matches what you'll see on every vendor's pricing page.

Pricing data comes from the MarginDash database: 410 models across 43 vendors, synced daily. Benchmark scores come from the Artificial Analysis Intelligence Index (AAII).

The standout Chinese models

These are the ten standout Chinese models scoring 39+ on the Intelligence Index — the threshold where models start matching Western flagship performance. Sorted by Intelligence Index, highest first.

The top scorer is Z AI's GLM-5 at 49.6 — higher than Claude 4.5 Sonnet (42.9) and GPT-5 medium (41.8). The mid-range has deepened: Alibaba now has two entries (Qwen3.5 397B at 45.0 and Qwen3.5 27B at 42.0), MiniMax has two (M2.5 at 42.0 and M2.1 at 39.5), and Xiaomi's MiMo-V2-Flash has crossed the flagship threshold at 41.4. On input pricing, these range from $0.10 to $1.00 per million tokens vs $1.25–$5.00 for Western flagships. On output, $0.30–$3.60 vs $10.00–$25.00. The gap is widest on output-heavy workloads.

Head-to-head: price comparison

Here's how the list prices compare side by side for models at similar Intelligence Index scores. All prices are per million tokens, published by each vendor.

The price gaps are stark. DeepSeek V3.2 charges $0.28/1M input and $0.42/1M output vs Claude 4.5 Sonnet's $3.00 and $15.00 — roughly 10x cheaper on input and 36x cheaper on output. The exact savings depend on your workload's input/output ratio.

At the top of the market, Kimi K2.5 matches Claude Opus 4.6 (Non-reasoning) on Intelligence Index (46.7 vs 46.4) while charging $0.60/$3.00 vs $5.00/$25.00. MiniMax-M2.5 matches GPT-5 medium (42.0 vs 41.8) at $0.30/$1.20 vs $1.25/$10.00 — a nearly identical capability swap at a fraction of the price.

GLM-5 beats Claude Sonnet 4.6 (Non-reasoning) on Intelligence Index — 49.6 vs 44.3 — at $1.00/$3.20 vs $3.00/$15.00. Alibaba's Qwen3.5 397B charges $0.60/$3.60 vs GPT-5.2's $1.75/$14.00, though with an Intelligence Index trade-off (45.0 vs 51.2).

Where Chinese models are strong

The benchmark data shows clear strengths:

• Advanced reasoning. Most of the flagship Chinese models are reasoning models. GLM-5 scores 49.6 on the Intelligence Index — higher than every Western model except Claude Opus 4.6 (53.0). Kimi K2.5 scores 46.7, ahead of GPT-5 (high) at 44.6. Qwen3.5 397B scores 45.0 with 89% on GPQA, and Qwen3.5 27B matches GPT-5 medium at 42.0 for just $0.30/1M input and $2.40/1M output.
• Graduate-level Q&A. Qwen3.5 397B hits 89% on GPQA, and Kimi K2.5 hits 88%, matching or exceeding most Western flagships. GLM-4.7 and Qwen3.5 27B both score 86%. These aren't watered-down models — they're competitive on the hardest public benchmarks.
• Price per token. On output pricing — where most of the cost lives — Chinese models are 4x to 36x cheaper depending on the pair. DeepSeek V3.2 charges $0.42/1M output vs Claude 4.5 Sonnet's $15.00. MiniMax-M2.5 charges $1.20/1M output vs GPT-5 medium's $10.00. Even GLM-5 at the top of the Chinese range ($3.20/1M output) undercuts Claude Sonnet 4.6 ($15.00) by nearly 5x.

Where to be cautious

Benchmarks measure capability. Running a model in production depends on more than that.

• Data residency. API calls to Chinese providers may route through infrastructure in China. For regulated industries or teams with data sovereignty requirements, this can be a non-starter. Check each vendor's data processing locations and terms before sending customer data.
• API stability and uptime. OpenAI and Anthropic have years of production API infrastructure and published SLAs. Chinese providers are newer to the global API market. Expect differences in rate limiting, error handling, and availability documentation.
• Reasoning overhead and latency. Eight of the ten models in the table above are reasoning models. They may have fast time-to-first-token, but total response time can be significantly longer because the model thinks before answering. For latency-sensitive applications like real-time chat, test actual end-to-end response times — not just benchmarks.
• Context window variation. DeepSeek V3.2 supports 128K context. Kimi K2.5 supports 256K. MiMo and MiniMax handle 128K–200K. If your use case involves long documents or extended conversations, check the context limit before committing — the cheapest model won't help if it can't fit your prompt.
• Ecosystem maturity. SDK support, function calling, structured outputs, streaming, and third-party integrations vary significantly. DeepSeek has the most mature developer ecosystem among Chinese labs. Others may require more integration work.
• Content filtering differences. Chinese models may have different content moderation policies. Some topics that work fine with Western APIs may hit filters, and vice versa. Test with your actual production prompts.

How to test a swap

Switching models based on benchmarks alone is a gamble. What you want is a side-by-side comparison on your actual workload — before you ship anything to production.

The MarginDash cost simulator lets you do this: pick a feature, select a Chinese alternative, and see projected cost savings based on your actual usage data. It filters out any model that drops more than 10% on benchmarks or can't handle your context window, so you're only comparing viable swaps.

If you're paying $3.00–$5.00/1M input and $10.00–$25.00/1M output for GPT-5 or Claude Sonnet and haven't looked at what MiniMax-M2.5, DeepSeek V3.2, GLM-5, or Qwen3.5 charge for comparable benchmark scores, you're likely leaving significant savings on the table.

You can explore all 410 models, filter by vendor, and run your own comparisons — sign up free to access the model database and cost simulator.

For value picks across all quality tiers (not just Chinese vendors), see The Cheapest AI Models That Are Actually Good.