GLM-5.3
Z.ai · Released 18 Aug 2026 · glm-5.3
$1.40
Input price / 1M tokens
#32 of 61, cheapest first
$4.40
Output price / 1M tokens
#27 of 61, cheapest first
1.05M
Context window
#11 of 71
70
Output tokens / second
#45 of 61
88.2%
Terminal-Bench 2.1
#5 of 17
66.9%
DeepSWE v1.1
#12 of 17
62.5%
Humanity’s Last Exam (with tools)
#7 of 16
Summary
GLM-5.3 is an open-weight model from Z.ai, released on 18 Aug 2026.
At $1.40 input and $4.40 output per million tokens, it is #32 of 61 on input price, cheapest first.
Its 1.05M-token context window ranks #11 of 71.
Measured output speed is 70 tokens per second, #45 of 61.
Z.ai's current flagship: same 744B-total / 40B-active base as GLM-5.2, improved only by post-training, with large gains in coding and cyber-security tasks. Weights license is MIT-style and allows commercial use; only a Model-as-a-Service operator with over $10B revenue in 12 months must pass a Z.ai security review first.
Benchmarks · 6 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| Terminal-Bench 2.1Claude Code harness, max effort | 88.2% | #5 of 17 | GLM-5.3 model cardSelf-reported | |
| Terminal-Bench 3.0Claude Code harness, max effort, avg@3 | 28.3% | – | GLM-5.3 model cardSelf-reported | |
| DeepSWE v1.1mini-swe-agent | 66.9% | #12 of 17 | GLM-5.3 model cardSelf-reported | |
| CyberGymClaude Code harness, max effort, Pass@1 | 84.5% | – | GLM-5.3 model cardSelf-reported | |
| Humanity’s Last Exam (with tools)with tools | 62.5% | #7 of 16 | GLM-5.3 model cardSelf-reported | |
| Toolathlon VerifiedPass@1, avg of 3 runs | 73% | – | GLM-5.3 model cardSelf-reported |
Self-reported means the lab ran the test itself. A rank appears only where several labs report the same version of a benchmark.
Similar models
Models with the most benchmarks in common with GLM-5.3, and the closest scores.
Terminal-Bench 2.1
#5 of 17Terminal tasks, earlier set
- GLM-5.388.2
- DeepSeek V4 Pro87.9
- Kimi K388.3
- DeepSeek V4.1 Flash90.6
- GPT-5.6 Terra87.4
- GPT-5.6 Luna84.7
DeepSWE v1.1
#12 of 17Long coding tasks in real repos
- GLM-5.366.9
- DeepSeek V4 Pro62.7
- Kimi K367.5
- DeepSeek V4.1 Flash74.2
- GPT-5.6 Terra69.6
- GPT-5.6 Luna67.2
Humanity’s Last Exam (with tools)
#7 of 16Expert questions, with search and code
- GLM-5.362.5
- DeepSeek V4 Pro60
- Kimi K356
- DeepSeek V4.1 Flash63.9
Compare GLM-5.3 side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- price: docs.z.ai/guides/overview/pricing, checked 2 Oct 2026
- context 1M, max output 128K, text-only input: docs.z.ai/guides/llm/glm-5.3, checked 2 Oct 2026
- release date: docs.z.ai/release-notes/new-released, checked 2 Oct 2026
- license, benchmarks: huggingface.co/zai-org/GLM-5.3, checked 2 Oct 2026
- parameters 744B-A40B: github.com/zai-org/GLM-5, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/glm-5-3, checked 2 Oct 2026
Building on GLM-5.3?
Get the architecture right before the bill arrives.
We size caching, routing and fallbacks for your workload on a free call, and tell you where a cheaper model is good enough.