Who Will Lead China’s Open‑Source LLM Race in 2025? A Deep Dive
The 2025 review reveals how Chinese open‑source large language models shifted from a single‑dominant player to a fierce top‑ten battle, highlighting DeepSeek R1’s breakout, the rise of Qwen, MiniMax, Kimi, upcoming IPOs, a detailed release schedule, and bold predictions for 2026.
2025 Open‑Source LLM Landscape in China
In 2024 the ecosystem largely depended on Llama 3, while Qwen2.5, Qwen Q, and DeepSeek V2/V2.5/V3 were recognized but remained niche. In 2025 DeepSeek’s R1 model went mainstream, prompting many Chinese companies to open their models.
Market Evolution
During the past twelve months the domestic open‑source LLM market transformed from “one or two dominant players” to a “top‑ten melee”. DeepSeek’s monopoly ended; now Qwen, GLM, MiniMax, Kimi and others each lead different segments, making a single‑player sweep impossible.
Key Players in 2025
Frontier: DeepSeek, Qwen
Close Competitors: Zhipu (Z.Ai), MiniMax, Kimi (Moonlight AI)
Worth Watching: Meituan Longcat, Tencent, StepFun, Baidu, ByteDance
Rising: Xiaomi, OpenBMB, InternLM, Skywork, Kuaishou, Baichuan, Huawei
2025 Major Open‑Source LLM Release Schedule
DeepSeek R1 – launched 20 January 2025; small team, MIT‑licensed, spurred many Chinese labs to open their models.
Qwen 3 – a comprehensive suite covering general, dense & MoE, vision, multimodal, coding, embedding, and reranking; Qwen 2.5 remains a “hidden gem”, while Qwen 3 has become the new multilingual default, surpassing Llama in downloads and fine‑tuning volume.
Kimi K2 – Moonlight AI’s focused single‑track model; its performance and distinctive style gained popularity.
MiniMax M2 – a dramatic upgrade from the modest M1, executed a flawless (Chinese‑style) launch and remains high on OpenRouter usage after the free period.
GLM‑4.5 – Zhipu’s breakthrough, introducing a widely‑liked lightweight Air version.
Outlook for 2026
2025 is the “year of legitimacy” for open‑source models – deployment is no longer merely possible, it is practical. Although top‑tier closed models still lead in robustness and richness, the gap has narrowed dramatically.
2026 Keywords Preview: multimodal unification / world models, edge inference, hybrid architectures, data openness, agent‑native / embodied intelligence.
2026 Predictions:
Open‑source models will continue scaling to larger sizes.
The security narrative around open models will see no substantive change.
The number of participants will keep growing.
General trends remain: MoE, hybrid attention, dense models for fine‑tuning.
On public benchmarks, the performance gap between open‑source and closed‑source frontier models will stay roughly constant.
Images and reference links illustrate the trends:
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.