Z.ai’s GLM-5.3 Achieves 60 on Intelligence Index, Competing with Kimi K3

Z.ai’s GLM-5.3 has been evaluated by Artificial Analysis, scoring 60 on its Intelligence Index as reported on August 18, 2026. This places the latest reasoning model from the Chinese lab on par with Moonshot AI’s Kimi K3, while trailing Anthropic’s Claude Opus 5, the current leader at 63.

Impressive Performance in Maximum Reasoning Mode

The score reflects GLM-5.3 operating at its maximum reasoning capability, a setting Z.ai recommends for coding tasks. At 60, it significantly surpasses the median score of 35 among 181 comparable models, ranking eighth overall.

Independent Evaluation Highlights Unique Model Construction

This evaluation serves as the first independent assessment of a model that Z.ai launched on August 14, 2026. Distinctively, GLM-5.3 utilizes the same base as GLM-5.2, with performance enhancements achieved through post-training rather than a complete pretraining reboot.

How GLM-5.3 Reached Its Performance Level

In Z.ai’s release announcement, the company detailed a month dedicated to scaling reinforcement learning in diverse, long-horizon task environments. This approach led to significant improvements in metrics like Terminal-Bench 3.0, which rose from 4.6 to 28.3, and DeepSWE v1.1, which climbed from 46.2 to 66.9. The model’s results span multiple categories, including coding, cybersecurity, and agentic benchmarks, competing against Kimi K3, Claude Opus 4.8, Claude Fable 5, and GPT-5.6 Sol. Detailed coverage of these features was provided by Unite.AI when the model was launched last week.

Understanding the Intelligence Index Evaluation

The score from Artificial Analysis carries significant weight as the evaluator personally conducts the evaluations. Their Intelligence Index v4.1.1 combines nine assessments across various areas, including agentic tasks, terminal coding, and scientific reasoning. GLM-5.3’s score reflects its performance across this broad spectrum, with a total evaluation cost of $1,238.50 on Z.ai’s API.

Competitive Standing Among Peers

GLM-5.3’s performance parity with Kimi K3 is noteworthy. Kimi K3, released July 16, 2026, also scores 60 on the index and retains its status as the top open-weights model according to Artificial Analysis. While it shares the score, GLM-5.3 remains proprietary, with 753 billion parameters recorded by Artificial Analysis.

Cost Efficiency Comparison with Other Models

When looking at pricing, GLM-5.3 is the more economical option. It costs $1.40 per million input tokens and $4.40 per million output tokens on Z.ai’s API, while Kimi K3 charges $3.00 and $15.00, respectively. In terms of task efficiency, GLM-5.3 achieves a cost of $0.68 per task compared to Kimi K3’s $0.84 and Claude Opus 5’s $2.34, although it generates a higher number of output tokens—170 million across the evaluation compared to the median of 72 million in its class.

Looking Ahead: What’s Next for GLM-5.3

GLM-5.3 is currently accessible through Z.ai’s API and has been rolled out to all GLM Coding Plan subscribers. Users must enable the thinking feature for optimal performance, with three adjustable effort levels. Z.ai cautions that applications without this feature enabled may fail until upgraded.

The release of model weights is pending. Z.ai has pledged to make them available two weeks after the August 14, 2026 launch, following necessary safety evaluations. This would position GLM-5.3 alongside Kimi K3 as an open-weight model at the 60 index score, approximately at half the per-token cost.