vs
10:4
overall capability wins (2 ties across 16 scored). GLM-5.3 leads overall.
- Reasoning4:1
- Coding3:1
- Arena1:0
- Agents0:1
- Tool use1:1
- Composite indices2:0
- Specs1:0
Verdict
GLM-5.3 leads coding (widest gap: +7.4 on SciCode); GLM-5.3 ranks higher on LMArena (+8 Elo).
- GLM-5.3 leads coding (widest gap: +7.4 on SciCode)
- GLM-5.3 ranks higher on LMArena (+8 Elo)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- GLM-5.3
- -
- GLM-5.3 Flash
- $0.28
Price per 1M tokens
Full comparison
Organization
Zhipu AI
Zhipu AI
Family
GLM
GLM
License
GLM-5.3 License
MIT
Open weights
Yes
Yes
Release
Aug 14, 2026
Aug 26, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
Zhipu / Z.ai
Modalities
text → text
text, image, video → text
Specs
Context window
1M
1M
Max output
128K
128K
Parameters
744B (40B act.)
320B (18B act.)
Pricing
Input $/1M
—
$0.15
Output $/1M
—
$0.50
Blended $/1M (3∶1)
-
$0.24
Speed
tok/s
-
49.8
TTFT (s)
-
1.51
Reasoning
MMLU-Pro
—
—
GPQA Diamond
91.7
91.2
Humanity's Last Exam
42.3
39.9
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
33.9
27.5
AA-LCR v1.1
79.7
80.0
CritPt
19.1
15.4
MMMU-Pro
—
—
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
83.9
84.3
Aider Polyglot
—
—
Terminal-Bench 3
28.3
—
BigCodeBench
—
—
SciCode
59.0
51.6
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
58.0
—
DeepSWE
66.9
63.0
WebDev Arena
1,614
1,607
Terminal-Bench 4.0
41.8
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,483
1,475
| Metric | GLM-5.3 | GLM-5.3 Flash | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Zhipu AI | Zhipu AI | — |
| Family | GLM | GLM | — |
| License | GLM-5.3 License | MIT | — |
| Open weights | Yes | Yes | — |
| Release | Aug 14, 2026 | Aug 26, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | Zhipu / Z.ai | — |
| Modalities | text → text | text, image, video → text | — |
| Specs | |||
| Context window | 1M | 1M | tie |
| Max output | 128K | 128K | tie |
| Parameters | 744B (40B act.) | 320B (18B act.) | A +424 |
| Pricing | |||
| Input $/1M | — | $0.15 | — |
| Output $/1M | — | $0.50 | — |
| Blended $/1M (3∶1) | - | $0.24 | — |
| Speed | |||
| tok/s | - | 49.8 | — |
| TTFT (s) | - | 1.51 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 91.7 | 91.2 | A +0.5 pts |
| Humanity's Last Exam | 42.3 | 39.9 | A +2.4 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 33.9 | 27.5 | A +6.4 pts |
| AA-LCR v1.1 | 79.7 | 80.0 | B +0.3 pts |
| CritPt | 19.1 | 15.4 | A +3.7 pts |
| MMMU-Pro | — | — | — |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 83.9 | 84.3 | B +0.4 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | 28.3 | — | — |
| BigCodeBench | — | — | — |
| SciCode | 59.0 | 51.6 | A +7.4 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | 58.0 | — | — |
| DeepSWE | 66.9 | 63.0 | A +3.9 pts |
| WebDev Arena | 1,614 | 1,607 | A +7 Elo |
| Terminal-Bench 4.0 | 41.8 | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,483 | 1,475 | A +8 Elo |
GLM-5.3
- Context
- 1M / 128K out
- Parameters
- 744B (40B act.)
- Price
- —
- Speed
- —
- Modalities
- text → text
- License
- GLM-5.3 License
Zhipu's Aug 14, 2026 flagship — GLM-5.2 base re-post-trained for complex software engineering, terminal, and real-world agent tasks, with SOTA open-weights coding and emergent cybersecurity skills (CyberGym 84.5). Open weights (744B MoE / 40B active) are on Hugging Face under the GLM-5.3 license; thinking effort is low/high/max (default max).
GLM-5.3 Flash
- Context
- 1M / 128K out
- Parameters
- 320B (18B act.)
- Price
- $0.15 / $0.50
- Speed
- 49.8 tok/s · 1.51s TTFT
- Modalities
- text, image, video → text
- License
- MIT
First natively multimodal GLM-5 (text/image/video in), MIT open weights, 320B total / 18B active MoE with 1M context, released Aug 26, 2026. List pricing $0.15/$0.50 per 1M (cache-hit ~$0.03), halved by a launch promo through Sept 9, 2026. Benchmark figures are Z.ai's self-reported launch numbers (DeepSWE 63.4, AutomationBench 48.8, Terminal-Bench 2.1 84.3); Artificial Analysis places it at 57 on its Intelligence Index.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.