GLM 5.2 vs GLM 5.3
| Benchmark | GLM 5.2 | GLM 5.3 |
|---|---|---|
| Humanity's Last Exam | 9.8 | 36.6 |
| Arena Text · overall | 1,476 · max | 1,479 · max |
| Arena Text · Chinese | 1,521.9 · max | 1,532.3 · max |
| Arena Text · English | 1,474.8 · max | 1,482.6 · max |
| Arena Text · French | 1,478.1 · max | 1,498.5 · max |
| Arena Text · German | 1,470.9 · max | 1,486.1 · max |
| Arena Text · instruction following | 1,465.9 · max | 1,480.3 · max |
| Arena Text · Japanese | 1,448.5 · max | 1,458.1 · max |
| Arena Text · Korean | 1,437.9 · max | 1,459 · max |
| Arena Text · non-English | 1,458.5 · max | 1,459.7 · max |
| Arena Text · Polish | 1,457.1 · max | 1,465.9 · max |
| Arena Text · Russian | 1,467 · max | 1,470.7 · max |
| Arena Text · Spanish | 1,477.6 · max | 1,461.3 · max |
| AA Omniscience Index | -6.6 | -8.4 |
| SimpleQA Verified v1 | 34.2 · max | 41 · max |
| CritPt (physics) | 3.1 | 14.6 |
| AA Intelligence Index 4.3.2 | 22.4 | 34.3 |
| AA Omniscience · Accuracy | 20.3 | 33.9 |
| AA Omniscience · Hallucination Rate | 33.7 | 64 |
| AA-LCR (long context) 1.1 | 42.3 | 72.7 |
| GDP.pdf · AA all-pass | 10.4 · max | 8.8 |
| MLCR-AA | 12.8 · max | 11.1 |
| Chess Puzzles | 22 · max | 37 · max |
| FrontierMath Tier 4 (v2) | 29.3 · max | 29.3 · max |
| FrontierMath Tiers 1–3 (v2) | 59.2 · max | 68.8 · max |
| Mystery Game Puzzles | 19 · low | 33 · max |
| Vals LegalBench | 84.1 | 84.8 |
| Vals MedCode | 40.8 | 42.9 |
| Vals MedScribe | 83.5 | 88.8 |
| Vals Legal Research Bench | 31.3 | 49 |
| IFBench (instructions) | 73.3 · max | — |
| ARC-AGI-1 · public evaluation | 80.4 | — |
| ARC-AGI-1 · semi-private | 77 | — |
| ARC-AGI-2 · public evaluation | 20.8 | — |
| ARC-AGI-2 · semi-private | 22.8 | — |
| EQ-Bench 3 | 1,563.2 | — |
| MathArena · Apex | 19.8 | — |
| MathArena · Apex Shortlist | 68.1 | — |
| MathArena · ArXivMath · April 2026 | 47.5 | — |
| MathArena · ArXivMath · February 2026 | 46.1 | — |
| MathArena · ArXivMath · June 2026 | 46.5 | — |
| MathArena · ArXivMath · March 2026 | 61.7 | — |
| MathArena · ArXivMath · May 2026 | 56.7 | — |
| MathArena · BrokenArXiv · April 2026 | 18.4 | — |
| MathArena · BrokenArXiv · February 2026 | 10.5 | — |
| MathArena · BrokenArXiv · June 2026 | 14.2 | — |
| MathArena · BrokenArXiv · March 2026 | 12.1 | — |
| MathArena · BrokenArXiv · May 2026 | 10.5 | — |
| MathArena · ArXivLean · June 2026 | 8.7 | — |
Specs
| Spec | GLM 5.2 | GLM 5.3 |
|---|---|---|
| Input, per 1M | !$1.40 | !$1.40 |
| Output, per 1M | !$4.40 | !$4.40 |
| Cost per task? | — | $0.85 |
| Speed, tok/s | 189 | 92 |
| Context | 1M | 1M |
| Weights | Open | Open |
| Released | Jun 2026 | Aug 2026 |
| Memory (4-bit) | 452 GB | 452 GB |
| Licence | MIT | glm-5.3 |
| Commercial use | Allowed | Restricted |
| See images | — | — |
| Read PDFs | — | — |
| Hear audio | — | — |
| Watch video | — | — |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | — |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ? |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
GLM 5.2 → GLM 5.3
Reasoning cannot be disabled›
Disabling reasoning is unsupported; keep reasoning enabled.
tool_choice auto · Z.ai API›
The first-party chat reference exposes only auto for tool_choice.
Complete unmodified reasoning_content›
Preserved thinking defaults on for Coding Plan and off for the standard API; when enabled, return complete unmodified reasoning_content in its original order to protect quality and cache hits.
Only one stop word›
Only one stop word is supported, so do not rely on a list of alternative stop sequences.
GLM 5.3 → GLM 5.2
Nothing known.