DevTk.AI
Chinese AI ModelsKimi K3GLM-5.3MiniMax M3DeepSeek V4.1AI API Pricing

Chinese AI Model API Prices 2026: GLM-5.3, Kimi K3, DeepSeek V4.1, Qwen 3.8

Compare current Chinese AI API prices across GLM-5.3, Kimi K3, DeepSeek V4.1, MiniMax M3, Qwen 3.8, and Xiaomi MiMo 2.6.

By Published Updated 4 min read

Chinese AI APIs are no longer one budget category. The current market includes ultra-cheap text routes, premium long-horizon coding models, native multimodal agents, and million-token context models.

The most useful question is not which model wins a vendor benchmark. It is which model completes your workload at the lowest total cost after output tokens, cached context, long-context premiums, latency, and retries.

Current Price Snapshot

USD prices below use official international API pricing. RMB prices remain in RMB because converting them would introduce exchange-rate drift.

ModelInput / 1MCached Input / 1MOutput / 1MContextBest Fit
DeepSeek V4.1 Flash, off-peak$0.15$0.003$0.601MCache-heavy long-output agents
DeepSeek V4.1 Flash, peak$0.30$0.006$1.201MConservative weekday price
Xiaomi MiMo-2.6 Flash$0.14$0.0028$0.281MAffordable multimodal agents
Xiaomi MiMo-2.6 Pro$0.435$0.0036$0.871MHarder multimodal agent work
Xiaomi MiMo-2.6 Pro UltraSpeed$4.35$0.036$8.701MLatency-sensitive work; no Batch
MiniMax M3, up to 512K input$0.30$0.06$1.201MLong-context multimodal agents
GLM-5.3 Flash$0.15$0.03$0.501MLow-cost GLM routing
GLM-5.3$1.40$0.26$4.401MFlagship coding and agent API
Kimi K3$3.00$0.30$15.001MMultimodal, very-long-output agents
Qwen3.8 Flash¥0.80¥0.10¥2.701MHigh-volume Alibaba Cloud work
Qwen3.8 Max¥12¥1.50¥361MPremium Alibaba Cloud model

MiniMax M3 input above 512K costs $0.60/M input, $0.12/M cached input, and $2.40/M output. Qwen3.8 Max cache writes cost ¥15/M; Qwen3.8 Flash cache writes cost ¥1.25/M. Xiaomi says MiMo-V2.5 will retire on October 21, 2026, so new integrations should use MiMo 2.6.

Kimi K3 automatic cache writes cost $3/M for the default five-minute TTL or $6/M for one hour; hits cost $0.30/M, refresh the TTL, and incur no additional write charge. MiniMax M3’s $0.30/$1.20 rate is a permanent 50% reduction; its current official table does not publish cache-write rates.

What Changed

Kimi K3 adds a 3T-class open-model candidate

Moonshot AI released Kimi K3 on July 16 with 2.8T total parameters, native visual understanding, 1M context, strict structured output, dynamic tool loading, and automatic prefix caching. The API currently always reasons at maximum effort, and the full model weights are now available.

At $3/$15, K3 is not the cheapest Chinese API. It should be evaluated where image/video input, very long output, or reduced retries can justify its premium. Moonshot also states that K3 still trails GPT-5.6 Sol and Claude Fable 5 overall.

MiniMax M3 makes long context affordable

MiniMax M3 launched on June 1, 2026 with a 1M-token context window, native image and video input, tool use, and a very large maximum output allowance. Its $0.30/$1.20 standard international price makes it one of the most interesting choices for long codebases and multimodal agents.

GLM-5.3 now has three priced API tiers

Z.AI now lists GLM-5.3 at $1.40/$4.40, GLM-5.3 Flash at $0.15/$0.50, and FlashX at $0.37/$1.25 for input/output. All three provide 1M context and 128K maximum output. This replaces the launch-period status where only Coding Plan access was available.

Qwen 3.8 and MiMo 2.6 replace August defaults

Qwen3.8 Flash and Max now offer 1M context and up to 131,072 output tokens through Alibaba Cloud Model Studio. Xiaomi’s MiMo 2.6 Pro and Flash keep the same unit prices as 2.5 while replacing the older endpoints for new deployments.

WorkloadStarting Model
Classification, extraction, simple tool callsGLM-5.3 Flash or MiMo-2.6 Flash
Cache-heavy coding agentDeepSeek V4.1 Flash
Long codebase, image/video input, long outputMiniMax M3
Multi-hour engineering workflowGLM-5.3
Image/video-grounded agent or output above 384KKimi K3
Existing Alibaba Cloud stackQwen3.8 Flash, escalate to Max

Do not route everything to one flagship. Measure task success rate, total generated tokens, retries, latency, and human repair time. A model with a higher token price can still be cheaper per completed task, but only your workload can prove that.

Use the AI Model Pricing Calculator to compare the USD-priced models in the canonical dataset.

Read the Kimi K3 API guide or the GLM-5.3 vs DeepSeek V4.1 vs Kimi K3 comparison.

Official sources checked October 2, 2026: Kimi K3 pricing, GLM-5.3 docs, Z.AI pricing, MiniMax pricing, Alibaba pricing, DeepSeek pricing, and Xiaomi MiMo pricing.

Next step

Related Posts