🤖 AI Tools
· 2 min read

MiniMax M2.7 vs GLM-5.1 vs Kimi K2.5 — Chinese Frontier Models Compared


Three Chinese AI labs are producing models that compete with Claude and GPT at a fraction of the cost. Here’s how they compare.

Quick comparison

MiniMax M2.7GLM-5.1Kimi K2.5
CompanyMiniMax (Shanghai)Z.ai (Beijing)Moonshot AI (Beijing)
Total params230B754B1T
Active params10B40B32B
SWE-Pro56.22%58.4%
Input price$0.30/1M$1.00/1M$0.60/1M
Speed100 tok/s55 tok/s
Context200K200K256K
LicenseOpen weightsMITMIT
Unique featureSelf-evolving8hr autonomousAgent Swarm
Training hardwareNVIDIAHuawei AscendNVIDIA

For coding

GLM-5.1 leads on SWE-Bench Pro (58.4%) and can work autonomously for 8 hours. Best raw coding quality of the three.

MiniMax M2.7 is the fastest (100 tok/s) and cheapest ($0.30/1M). Best value for routine coding.

Kimi K2.5 has Agent Swarm for parallel execution. Best for parallelizable tasks like batch refactoring.

For the race

All three are represented in our AI Startup Race:

  • GLM runs through Claude Code with the Z.ai Coding Plan
  • Kimi runs through Kimi CLI
  • MiniMax is not in the race (yet) but could be a strong contender

Which to pick

Cheapest: MiniMax M2.7 ($0.30/1M) Best quality: GLM-5.1 (58.4% SWE-Pro) Best for parallel work: Kimi K2.5 (Agent Swarm) Best for GDPR: None — all are China-based. Use Mistral for EU compliance.

Or use all three via OpenRouter and model routing.

The bigger picture

Chinese AI labs are now producing 4 of the top 10 coding models globally. A year ago, the conversation was “Claude vs GPT.” Now it’s “which of the 6 frontier-class models gives the best value?”

For developers, this is great — competition drives prices down and quality up. MiniMax at $0.30/1M, DeepSeek at $0.27/1M, and GLM’s Coding Plan at $18/month mean frontier-class AI coding is accessible to everyone, not just well-funded teams.

The models in our AI Startup Race represent this diversity: 3 US models (Claude, GPT, Gemini) competing against 4 Chinese models (DeepSeek, Kimi, Xiaomi, GLM). The race will show whether the cheaper Chinese models can match the more expensive US ones in real-world startup building.

Related: Best Open-Source Coding Models 2026 · GLM-5.1 vs DeepSeek vs Qwen · How to Reduce LLM API Costs