GLM 5.2 in the catalog: a million tokens and reasoning from Z.AI

We have added GLM 5.2 (glm-5.2), the senior model of the Z.AI lineup. Compared to GLM 4.x this is a big step: a million tokens of context, built-in reasoning and careful tool use.

What matters in practice

• 1M context. This is the size at which you no longer need to slice documentation into pieces. A whole mid-sized monorepo, a book of requirements, an annual report: it all fits.
• Reasoning on demand. The model can reason before answering, and on simple questions it does not waste extra tokens on it. On olympiad-level math and logic puzzles this noticeably raises accuracy.
• Tools. Function calling works predictably, without inventing non-existent arguments, which is exactly where early GLM models stumbled.

How to call it

curl https://api.yenisei.ru/v1/chat/completions \
  -H "Authorization: Bearer $YENISEI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "messages": [
      {"role": "user", "content": "Here are 400 pages of requirements. Find contradictions in the authorization requirements."}
    ]
  }'

Russian language

GLM has historically been oriented toward Chinese and English, and that showed in earlier versions: Russian was "understandable but wooden". In 5.2 it has gotten noticeably better: the bureaucratic tone is gone and terminology holds up. For internal analytics and document work it is fine. For marketing copy that goes to a client without edits, we would still run it through an editor.

When to pick it, and when not to

GLM 5.2 is worth choosing when you need a large context and reasoning at the same time: analyzing long legal or technical documents, working through large logs with conclusions, complex research. If the task is short chatbot replies or classification, that is overkill: take GLM 5.3 Flash (glm-5.3-flash) or GLM 4.7 Flash (glm-4.7-flash). And if you need the strongest model of the lineup, look at GLM 5.3 (glm-5.3): it is the next release with an even longer window.

The model is already in the catalog; access requires no separate approval.