Z.ai Ships GLM-5.2: Usable 1M-Token Context, No Benchmarks

One-liner: Z.ai launched GLM-5.2 on June 13 with a functional 1-million-token context window — five times its predecessor — but made the unusual choice to ship without any public benchmark numbers.

Key Facts

• 744B-parameter Mixture-of-Experts architecture; 40B active parameters per token
• 1M-token context (≈5× GLM 5.1's ~200K limit); output up to 131,072 tokens
• Two reasoning modes: High and Max thinking effort
• API pricing: $1.40 input / $4.40 output per million tokens
• No benchmark scores at launch — no SWE-bench, LiveCodeBench, or HumanEval published
• MIT-licensed open weights promised via Hugging Face in the week following launch

Why It Matters

The benchmark-free launch is the real story. Rather than competing on leaderboard rankings, Z.ai led with raw context capacity for coding and agentic workflows. If the 1M-token window delivers in production, it shifts competitive pressure from eval metrics toward real-world throughput — a pattern worth watching as Chinese labs increasingly skip the benchmark arms race in favor of capability-first releases.

Read More

• Z.ai Launches GLM-5.2 — MarkTechPost