IntermediateAI

GLM-5.1: Z.ai’s Open-Weight Model Takes #1 on SWE-Bench Pro

On April 7, 2026, Z.ai (formerly Zhipu AI) released GLM-5.1 — a 754-billion-parameter open-weight Mixture-of-Experts model that claims the #1 position on SWE-Bench Pro with a score of 58.4%, edging out Claude Opus 4.6 (57.3%). Licensed under MIT and designed for agentic engineering, GLM-5.1 can autonomously sustain coding tasks for up to eight hours across hundreds of iterations.

GLM-5.1 benchmark comparison chart showing performance across coding, math, and agentic tasks
Image credit: Z.ai / HuggingFace

What’s New in GLM-5.1

GLM-5.1 is a post-training upgrade to GLM-5, built on a Dynamic Sparse Attention (DSA) MoE architecture with approximately 40 billion active parameters per token. The model is laser-focused on two areas: coding and agentic tool use.

On coding benchmarks, GLM-5.1 leads across multiple evaluations:

  • SWE-Bench Pro: 58.4% — the highest score among all models, open or closed
  • CyberGym: 68.7% — top score for cybersecurity tasks
  • BrowseComp: 68.0% — best in web browsing comprehension
  • Terminal-Bench 2.0: 63.5% on the Terminus-2 track
  • NL2Repo: 42.7% for natural-language-to-repository generation

Math and reasoning capabilities remain strong: 95.3% on AIME 2026, 86.2% on GPQA-Diamond, and 83.8% on IMOAnswerBench.

Agentic Engineering at Scale

What sets GLM-5.1 apart is its ability to sustain long-horizon agentic tasks. In demonstrations, the model autonomously built a complete Linux desktop system over an eight-hour session, performing 655 iterations of planning, execution, testing, and optimization. In another test, it increased vector database query throughput to 6.9× the initial production version through iterative experimentation.

The model supports deployment through SGLang (v0.5.10+), vLLM (v0.19.0+), xLLM, Transformers, and KTransformers. Its MIT license places no restrictions on commercial use.

What This Means

GLM-5.1 represents a milestone for Chinese open-source AI: a model from Z.ai that outperforms the best closed-source competitors on SWE-Bench Pro, the industry’s most respected coding benchmark. For developers and researchers, the MIT license and broad framework support make it immediately deployable — no API key required.

This post was drafted with AI assistance and reviewed by RITS staff.

Sources

Related stories

agentic ai

GLM-5: Zhipu AI Ships a 744B Open-Weight Frontier Model

Key points On February 11, 2026, Zhipu AI released GLM-5 through its Z.ai platform with 744 billion total parameters and a Mixture-of-Experts architecture. GLM-5 has 744B total parameters with 40B active per token, spanning 80 layers and 256 experts of which 8 are active at inference, a sparsity rate of about 5.9%. GLM-5 scored 50 […]

agentic ai

GLM-4.7-Flash: Z.ai’s Efficient 30B MoE Model for Coding and Agents

Key points Z.ai released GLM-4.7-Flash on January 20, 2026, a fully open-source model under the MIT license that is available for free via the Z.ai API platform. Despite its 30B label, the model uses a Mixture-of-Experts architecture with 31.2 billion total parameters but only approximately 3 billion active parameters during any given inference pass. It […]

AI

GLM-OCR: Z.ai’s 0.9B Model Takes the Top Spot on Document Understanding Benchmarks

Key points Z.ai has open-sourced GLM-OCR, a compact multimodal model with 0.9 billion parameters that ranks first on OmniDocBench V1.5 with a score of 94.62. GLM-OCR is a multimodal optical character recognition model developed by Z.ai, the commercial arm of ZhipuAI, and designed specifically for complex document understanding. It supports three core recognition modes: general […]

Never miss a RITS update

Subscribe to the RSS feed for AI, emerging technology, and digital scholarship news curated for the NYU Shanghai community, or get in touch to suggest a story.