GLM-5.2 Released with 1M Token Context and 753B Parameters Under MIT License
Zhipu AI has released GLM-5.2, a 753 billion parameter model featuring a 1 million token context window and MIT open-source license. The model scores 62.1% on SWE-bench Pro and 91.2% on GPQA-Diamond, with flexible reasoning effort levels for coding tasks.
GLM-5.2 — Quick Specs
GLM-5.2 Released with 1M Token Context and 753B Parameters Under MIT License
Zhipu AI has released GLM-5.2, a 753 billion parameter model that delivers a 1 million token context window under an MIT open-source license. The model represents an architectural shift with its IndexShare feature, which reduces per-token FLOPs by 2.9× at 1M context length by reusing the same indexer across every four sparse attention layers.
Benchmark Performance
According to Zhipu AI, GLM-5.2 achieves:
- SWE-bench Pro: 62.1% (versus 58.4% for GLM-5.1)
- GPQA-Diamond: 91.2% (versus 86.2% for GLM-5.1)
- NL2Repo: 48.9% (versus 42.7% for GLM-5.1)
- DeepSWE: 46.2% (versus 18% for GLM-5.1)
- AIME 2026: 99.2%
- HLE reasoning: 40.5% (54.7% with tools)
The company positions GLM-5.2 competitively against Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro on long-horizon task benchmarks, though these comparisons represent company claims rather than independent verification.
Technical Architecture
GLM-5.2 introduces two key architectural improvements:
- IndexShare: Reuses indexers across sparse attention layers, reducing computational requirements at extended context lengths
- Enhanced MTP layer: Improved for speculative decoding, increasing acceptance length by up to 20%
The model supports multiple "thinking effort levels" for coding tasks, allowing developers to balance performance against latency requirements.
Pricing and Availability
Pricing not yet disclosed. The model is available through:
- Z.ai API Platform for hosted inference
- Local deployment via SGLang (v0.5.13.post1+), vLLM (v0.23.0+), xLLM (v0.10.0+), Transformers (v0.5.12+), and KTransformers (v0.5.12+)
- Hugging Face model hub (zai-org/GLM-5.2)
What This Means
GLM-5.2's MIT license removes geographic restrictions common in other frontier models, potentially accelerating adoption in regions where licensing has been restrictive. The architectural focus on computational efficiency at extreme context lengths—2.9× reduction in FLOPs—addresses a critical bottleneck as context windows expand industry-wide. However, the benchmark scores show GLM-5.2 trailing Claude Opus 4.8 and GPT-5.5 on most coding benchmarks, with particular gaps on SWE-bench Pro (69.2% vs 62.1%) and NL2Repo (69.7% vs 48.9%), suggesting it competes more directly with open-weight alternatives than proprietary frontier models.
Related Articles
Tencent Open-Sources AuK, a 1.5B-Parameter Speech Generation and Editing Model
Tencent has open-sourced AuK, a 1.5B-parameter foundation model for speech generation and editing that handles TTS, content editing, and audio enhancement through natural-language instructions. The release includes a distilled AuK-Flash variant for 4-step fast inference, both under MIT license.
AllSpark's Iris-mini and Iris-pro Top Open-Weight Search Agent Benchmarks
Chinese lab AllSpark has released Iris-mini and Iris-pro, two open-weight search agents built on Qwen3 models that claim the top spot among open-weight systems in their size classes on four research benchmarks. The release includes model weights, an agent harness, and evaluation code, with training pipelines to follow.
Google Releases TimesFM-3, a 330M-Parameter Model That Forecasts Sales Using Weather and Discount Data
Google Research has released TimesFM-3, a 330-million-parameter time series forecasting model that predicts outcomes like sales by combining related variables, historical data, and known future events such as discounts or weather. The model claims top rankings on three benchmarks against Amazon's Chronos-2 and the Toto-2.0 family.
DeepSeek Ships V4.1-Flash With Novel Encoder-Decoder Architecture, Cuts KV Cache to 1/8 of Predecessor
DeepSeek released V4.1-Flash, a 763B-parameter model built on a new causal encoder-decoder architecture that splits 8B active parameters for prefill and 16B for decode. The model adds native vision support, a 1M-token context window, and shrinks KV cache footprint to roughly 1/8 of DeepSeek V4 Flash, while retiring V4 Pro.
Comments
Loading...