Z Ai: GLM 5.2
Open-weight coding model with 99.2 on AIME 2026, 91.2 on GPQA-Diamond, and a ~1024K token context window. Priced at $1.40 input / $4.40 output per million tokens, it rivals closed models at a fraction of the cost. Text-only, weights available on Hugging Face.
Specifications
| Attribute | Value |
|---|---|
| Lab | Z Ai |
| Tags | Intelligent Coding Agentic Long Context |
| Release Date | 2026-06 |
| Context Window | 1,048,576 tokens |
| Input Price / 1M | $1.40 |
| Output Price / 1M | $4.40 |
| Input Modalities | Text |
| Output Modalities | Text |
Strengths
- 99.2 on AIME 2026 (vendor-reported)
- 91.2 on GPQA-Diamond (vendor-reported)
- 62.1 on SWE-bench Pro (vendor-reported)
- ~1024K token context window
- Open-weight MIT license
Weaknesses
- Text-only, no multimodal support
- High verbosity compared to peers
- Long-context accuracy degrades on very large inputs
- Limited independent verification as of June 2026
Best For
- Long-horizon coding projects
- Agentic workflows with extended context
- Large document analysis and scientific reasoning
- Post-training and fine-tuning tasks
In Depth: GLM 5.2
Benchmark Performance
GLM 5.2 (#13), placing it ahead of GPT-5.3 (rank #8, 9.4/10) in value but behind Claude Opus 4.8 (rank #2, 9.7/10) in raw performance. It leads all open-weights models on the Artificial Analysis Intelligence Index v4.1 with a score of 51.
GLM 5.2 scores 99.2 on AIME 2026, 91.2 on GPQA-Diamond, 62.1 on SWE-bench Pro, and 74.4 on FrontierSWE (Dominance), per vendor-reported results. The Artificial Analysis Intelligence Index v4.1 rates it 51, the highest of any open-weights model. It excels in coding benchmarks but trails Claude Opus 4.8 (rank #2, 9.7/10) in SWE-Marathon and long-context tasks.
Pricing & Value
GLM 5.2 costs $1.40 per million input tokens and $4.40 per million output tokens, with cached input at $0.26 per million tokens. This is 7x cheaper than GPT-5.5 Pro ($30.00 input) and 11x cheaper than Claude Opus 4.5 ($15.00 input).
At $1.40 input and $4.40 output per million tokens, GLM 5.2 delivers 9.2 LMRank points per dollar at a cost of $0.31 per million tokens, versus Claude Sonnet 5 (rank #6, 9.4/10) at $0.79 per million tokens. This makes GLM 5.2 roughly 2.5x more cost-efficient per LMRank point than its Anthropic peer.
Who Should Use This
GLM 5.2 is for developers building coding agents, researchers analyzing large documents, and cost-conscious teams wanting open-weight performance. It is not for multimodal tasks.
- Developers: Build long-horizon coding agents with ~1024K context and 128K output tokens tradeoff is higher verbosity.
- Researchers: Analyze large documents and scientific reasoning tasks but text-only, no multimodal support.
- Cost-conscious teams: Get open-weight performance at $1.40 input / $4.40 output tradeoff is lower long-context accuracy vs Claude Opus 4.8.
- Not for: Teams needing multimodal input or image generation GLM 5.2 is text-only.
Release & Version History
GLM 5.2, released June 2026 by Z Ai (Zhipu AI, Beijing), advances the GLM-5 family with a ~1024K token context window and open-weight licensing under MIT. It is the first model in the series to achieve state-of-the-art scores among open-weights models.
Released June 16, 2026, GLM 5.2 is the successor to GLM 5.1 within the GLM-5 family. It uses a Mixture of Experts (MoE) architecture with 753 billion total parameters (~40 billion active), featuring DSA (Dense-Sparse-Alternating) attention with IndexShare for long-context efficiency. It is licensed under MIT open-source with weights available on Hugging Face and ModelScope.