GLM-5.3
Category scores
Scores per capability area with the model's rank among ranked models in that category.
- 1Reasoning#1179.8
- 2Agentic & tool use#874.5
- 3Math71.5
- 4Knowledge#4464.8
- 5Coding#2958.6
Benchmark results (26)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| Terminal-Bench 2.1 | 88.2 | #8 of 68 |
| MazeBench | 0 | #20 of 42 |
| HLE w/ tools | 62.5 | #5 of 27 |
| CyberGym | 84.5 | #7 of 25 |
| AutomationBench | 48.2 | #11 of 25 |
| Toolathlon-Verified | 73.0 | #13 of 22 |
| Agents' Last Exam | 28.5 | #11 of 17 |
| Hermes Index | 39.2 | #7 of 16 |
| Hermes Bench | 71.8 | #7 of 16 |
| ExploitGym | 15.0 | #8 of 14 |
Coding
| DeepSWE | 66.9 | #27 of 50 |
| NL2Repo | 58.0 | #6 of 31 |
| Bug Hunt Bench | 19.0 | #19 of 23 |
| FrontierSWE v2 | 30.2 | #10 of 20 |
| ProgramBench | 19.0 | #14 of 15 |
| VulcanBench v3 | 78.3 | #12 of 14 |
| OpenHarmony Bench | 60.8 | #1 of 12 |
| FrontierSWE | 78.1 | #2 of 3 |
| PostTrain Bench | 39.8 | #1 of 3 |
Math
| FrontierMath v2 (Tiers 1-3) | 68.772 | #24 of 64 |
| FrontierMath v2 (Tier 4) | 29.268 | #27 of 51 |
Multimodal
| Design Arena Website | 1295 | #17 of 91 |
Instruction following
| Gray Swan IPI (15 attempts)(lower is better) | 31.5 | #10 of 12 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| List price (OpenRouter) | $0.079 | $1.40 | — |
| OpenRouter z-ai/glm-5.3 | $0.079 | $1.40 | $0.078 |
Z.AI publishes the GLM-5.3 FP8 checkpoint under the custom GLM-5.3 License for self-hosting and does not publish a distinct first-party hosted per-token rate for this exact checkpoint. We represent the open-weight row as self-host/free-per-token before infrastructure costs. The hosted Z.AI route and Coding Plan must not be read as free from this self-host placeholder.
Price history
- Oct 17, 2026: $0.079 / $4.80
- Oct 19, 2026: $0.079 / $1.40
Arena ratings
Human-preference ratings from LMArena. Best-rated variant shown.
| Text(max) | 1472 | #27 · 22.5K votes |
| WebDev (coding)(Max) | 1621 | #25 · 9.8K votes |
Open weights
From the Hugging Face repository.
- Parameters
- 753B
- Licence
- other
- Downloads (30 days)
- 1.9M
- Repository
- zai-org/GLM-5.3 ↗
Speed
Measured by us through OpenRouter (median of trials).
Not measured yet.
Related models
Same family
- GLM-5.260.5
- GLM-5.156.3
- GLM-554.9
- GLM-5-Turbo53.2
- GLM-5V-Turbo49.6
Similar score
- Ling 3.1 Flash68.1 · $0.11
- Claude Haiku 5.567.2 · $0.20
- Claude Opus 4.7 (Adaptive)68.7 · $10.00
- GPT-5.4 Pro68.7 · $67.50
- MiniMax M3.1 Flash67.1 · $0.26