Context up to a million tokens
Versus 5.1’s ~200K, long specs and large repos can stay in one conversation.
Zhipu · GLM 5.2 · Jun 2026
GLM 5.2 — long-task coding with million-token context
- Million-token context
- Very long single outputs
- Tunable Thinking
- Open weights to compare
What is GLM 5.2?
GLM 5.2 is Zhipu’s open model released in June 2026, built to push long-task coding and terminal tool collaboration inside a million-token context. Versus GLM 5.1, context rises to a million tokens, with stronger long-context efficiency and a tunable thinking budget. For vision multimodal needs, use the matching vision model—this page covers text capability. Pick GLM 5.2 in iMini Agent to use it.
- Vendor
- Zhipu
- Released
- Jun 2026
- Context
- 1M tokens
- Max output
- 128K tokens
- Deep thinking
- High / Max tunable
- Best for
- Long-task coding & terminal tools
What’s new vs GLM 5.1
Context, long-thread efficiency, and thinking budget—the changes to check before you keep using 5.2.
Cheaper long context
Official materials stress designs like IndexShare that cut million-token cost—better for steady long runs.
Tunable Thinking depth
Open High / Max on hard jobs; lower the budget on simple ones for speed.
Stronger open coding results
Official headline benches like Terminal-Bench and SWE-Bench Pro improve vs the prior gen—fit as an engineering workhorse.
Official benchmarks
Charts from Zhipu’s GLM-5 series resources—coding, long-horizon tasks, and Vending Bench. All models at maximum thinking strength.
Three common workflows
GLM 5.2 fits bilingual long-task coding and terminal tool collaboration.
Long-horizon coding
Cross-file features, refactors, and test fixes. Raise thinking on hard jobs.
Terminal and tool collaboration
CLI environments and multi-step tools. Pin success checks and stop conditions.
CN/EN knowledge work
Research synthesis and decision memos—for China delivery and bilingual materials.
Vs GLM 5.1 and Claude Opus 4.8
All three lean long-horizon engineering. Gaps are generation, open stack, and Claude’s ceiling.
| Dimension | GLM 5.2 | GLM 5.1 | Claude Opus 4.8 |
|---|---|---|---|
| Context | 1M tokens | ~200K tokens | 1M tokens |
| Reasoning | High / Max tunable | Built for long runs | Usually high depth |
| Coding / Agents | Long-task coding & terminal tools | Long-task engineering | Opus coding collaboration |
| Long tool runs | Long-thread delivery | Hours of autonomous work | Steady Opus collaboration |
| Speed posture | Tunable thinking budget | Heavier long runs | Quality and experience balanced |
| Prefer when | New Zhipu long-task coding | Workflows still on 5.1 | Tied to the Claude stack |
Using GLM 5.2 on iMini
Free quota, model identity, key specs, and how it differs from 5.1.
Use GLM 5.2 free on iMini Agent
Million-token context and tunable Thinking—for long-task coding and tool collaboration.
Open GLM 5.2No Zhipu API key required.