TL;DR
Chinese laboratories released four frontier-class open-weight models between April 24 and mid-June 2026. The pace could lower self-hosting costs and narrow the gap with closed systems, but benchmark comparisons, licensing durability and regulatory exposure remain unsettled.
Chinese AI laboratories released four frontier-class open-weight models in roughly eight weeks, from DeepSeek V4 on April 24 to new Kimi and GLM systems in mid-June 2026. The compressed release schedule matters because it could accelerate improvements in lower-cost, self-hosted AI while shifting more of the open-model market toward Chinese developers.
The sequence began with DeepSeek V4 Pro and Flash, followed by MiniMax M3 on June 1. Moonshot AI released Kimi K2.7-Code around June 13, while Z.ai introduced GLM-5.2 within the same mid-June period, according to the release timeline compiled by Thorsten Meyer AI.
The source describes the systems as downloadable open-weight models, with most offered under MIT or similar permissive terms. DeepSeek V4 is listed as a mixture-of-experts model with 1.6 trillion total parameters, 49 billion active parameters per pass and a one-million-token context window. MiniMax M3 is described as supporting native multimodal use and the same context length under a modified MIT license.
Moonshot positions Kimi K2.7-Code as an agent-focused coding model and says it uses about 30% fewer reasoning tokens than K2.6. Z.ai’s GLM-5.2 is listed as a 753-billion-parameter mixture-of-experts system. Those efficiency and performance statements come from the developers or the supplied analysis; they are not equivalent to independently reproduced findings.
Four Frontier-Class Open Models in Eight Weeks
China’s Release Cadence Is the Story
Same-day-verified market pulse · July 13, 2026
The production line — spring 2026
The board this week — BenchLM overall score, July 2026
Gift & complication — the European read
The gift
Frontier-adjacent capability, permissive licenses, weeks-long refresh cycle. This cadence is what makes serious on-premises AI economically thinkable in 2026.
The complication
Still a dependency — geopolitical, not technical. Hosted Chinese APIs fall under Chinese data law; many Western agencies won’t touch the weights at all. Licensing generosity is a policy, not a law of nature.
The signal: if your infrastructure strategy assumes open models improve slowly, it’s already wrong. If it assumes the current licensing generosity is permanent, it’s unhedged.
Open Models Accelerate Their Cycle
The main development is the frequency of high-end releases, not any single model. A weeks-long cycle gives companies running models on their own infrastructure more frequent opportunities to improve capability, context length and operating cost without moving workloads to a closed Western service.
The July 2026 BenchLM snapshot cited by Thorsten Meyer AI gives DeepSeek V4 Pro a score of 87, compared with 93 for the leading proprietary model. The same table places GLM-5.1 at 83, Kimi K2.6 at 81 and Qwen 3.5 397B at 79. That suggests several Chinese model families are competing near the upper end, although one composite benchmark cannot establish performance across every workload.
The source also estimates hosted access to these models at five to 30 times cheaper than Western frontier APIs. If that comparison holds for real workloads, it could alter spending decisions for developers, European companies and public bodies seeking local-first or sovereign AI. Actual costs depend on token use, hardware, latency and the amount of engineering needed to operate a model safely.
Server Room Temperature and Humidity Monitor for Data Centers,Pharmaceuticals Alongwith Factory Calibration Certificate Model: AI-RHTx-IOT (RHTx-IoT Hosting to Customer End (Without Hosting))
- Model and Connectivity: RHTx-IoT with SMS, Email, Cloud
- Measured Parameters: Temperature and Humidity
- Temperature Range and Accuracy: 0-50°C, ±0.5°C
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
China Gains Open-Model Depth
China’s open-weight market now includes strong offerings from DeepSeek, Z.ai, Moonshot AI and Alibaba. Their strategies differ: DeepSeek emphasizes low prices, Moonshot targets long agent runs, Z.ai competes on benchmark performance, and Alibaba’s Qwen family offers models across several hardware tiers, including smaller variants suited to local deployment.
Thorsten Meyer AI says four of the five strongest open-weight families now come from Chinese laboratories. It contrasts that breadth with a thinner Western field, while identifying Ai2’s Olmo line as a more fully open alternative. The classification depends on how trackers define open source versus open weight, since downloadable parameters do not always include training data, complete code or unrestricted terms.
HIWONDER Robot Car with ChatGPT Large AI Models, 3D Depth Camera Ackermann Chassis ROS2-HUMBLE Lidar SLAM Mapping Navigation Autonomous Driving, MentorPi A1 Standard Kit with Raspberry Pi5 8GB
- Compatible with Raspberry Pi 5: Supports Raspberry Pi 5 and ROS2
- High-Performance Hardware: Ackerman chassis, lidar, depth camera, motors
- Advanced AI Features: SLAM, path planning, vision recognition, tracking
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Benchmarks and Licenses Remain Unsettled
It is not yet clear whether the eight-week release pace can continue or whether every model will perform at the reported level in production. The supplied BenchLM results are a single July 2026 snapshot, and the source itself cautions against treating one tracker as definitive. Independent testing across coding, agent reliability, multilingual work and safety could produce a different ordering.
The material also contains a version mismatch: its BenchLM table lists GLM-5.1 at 83, while the release timeline and Artificial Analysis claim refer to GLM-5.2. No directly comparable GLM-5.2 BenchLM score is supplied. Future license terms, Chinese export policy and the long-term availability of permissively licensed weights also remain unknown.
ESP32 Basic Starter Ai Chatbot Kit Development Board USB-C Dual Core Microcontroller Support AP/STA/AP+STA Compatible with Arduino IDE IoT for Beginners Engineers
- High-Performance Dual-Core CPU: Equipped with dual-core processor and Type-C USB
- Rich Peripheral Support: Includes SPI, LCD, Camera, UART, I2C, and more
- Wireless Connectivity: Supports Wi-Fi 2.4GHz and Bluetooth 5 (LE)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Tests Will Settle Claims
The next test will be whether independent benchmark providers reproduce the reported gains and whether users see comparable results in live deployments. Pricing, throughput, hardware requirements and agent stability over long tasks will matter as much as composite benchmark scores.
European and US organizations will also have to distinguish between self-hosting downloaded weights and sending prompts to a hosted Chinese API. Procurement reviews, data-location rules and future model licenses will shape adoption. The next round of releases will show whether the spring 2026 sequence was a durable production cycle or a temporary cluster.
Coding with AI For Dummies (For Dummies: Learning Made Easy)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Which four models were released?
The sequence comprised DeepSeek V4, MiniMax M3, Moonshot AI’s Kimi K2.7-Code and Z.ai’s GLM-5.2 between April 24 and mid-June 2026.
Are these models open source?
They are described as open-weight and downloadable, mostly under permissive or modified MIT-style licenses. Open weight does not always mean fully open source, because training data and the complete development process may remain private.
Are Chinese open models now equal to closed leaders?
Not on the supplied BenchLM composite. DeepSeek V4 Pro scored 87, six points behind the proprietary leader at 93. The difference may vary by task, and one benchmark cannot settle overall capability.
Why are European organizations interested?
Permissive weights can support local deployment and lower costs, reducing reliance on external API providers. Chinese-hosted services may still create data-governance and procurement barriers for regulated workloads.
Can this release pace continue?
That remains unknown. The four-model cluster shows that Chinese laboratories can release high-end systems rapidly, but hardware access, development costs, licensing policy and export rules could affect future release schedules.
Source: Thorsten Meyer AI
