GLM 5.3 vs Kimi K3
| Benchmark | GLM 5.3 | Kimi K3 |
|---|---|---|
| AutomationBench-AA | 52 | 58.3 |
| GDPval-AA (work tasks) 2.1 | 1,334 | 1,534.1 |
| τ³-Banking · AA | 50.3 · max | 46 |
| AA Analyst Agent | 38.8 · max | 38.8 |
| ITBench SRE | 46.1 · max | 47.7 |
| Terminal-Bench 4.0 | 34.8 | 12.6 |
| Terminal-Bench Science 0.1 | 3.8 | 8.1 |
| AA-Briefcase 1.1 · Elo | 1,184.4 | 1,498.8 |
| AA-LCR (long context) 1.1 | 72.7 | 88.7 |
| DeepSWE 1.1 | 69 · max | 68.5 |
| EnterpriseOps-Gym-AA | 36.4 · max | 45.3 |
| FrontierCode 1.1 | 34.9 | 43.6 |
| Harvey LAB-AA 1.1 | 0.3 · max | 5.3 |
| SWE-bench Pro (v2 full) | 95.6 · max | 97.7 |
| SWE-bench Pro (v2 hard) | 84.3 · max | 88.2 |
| Toolathlon-Verified | — | 76.5 |
| τ³-bench · banking · 1.0.1 | — | 37.1 |
| APEX-Agents | — | 41.3 |
| Furniture Assembly | — | 34.2 |
Specs
| Spec | GLM 5.3 | Kimi K3 |
|---|---|---|
| Input, per 1M | !$1.40 | $3 |
| Output, per 1M | !$4.40 | $15 |
| Cost per task? | $0.85 | $2.00 |
| Speed, tok/s | 92 | 55 |
| Context | 1M | 1M |
| Weights | Open | Open |
| Released | Aug 2026 | Jul 2026 |
| Memory (4-bit) | 452 GB | 1,668 GB |
| Licence | glm-5.3 | kimi-k3 |
| Commercial use | Restricted | Restricted |
| See images | — | ✓ |
| Read PDFs | — | Partial |
| Hear audio | — | — |
| Watch video | — | ✓ |
| Call tools | ✓ | ✓ |
| Force a tool call | — | ✓ |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | Partial |
| Run code | ? | ? |
| Use a computer | ? | ? |
What breaks when you switch
Memory (4-bit): 452 GB → 1,668 GB
Sampling parameters are fixed›
temperature, top_p, n and penalty values are fixed; omit them from requests.
No public image URLs›
Vision input rejects public image URLs; use base64 or ms:// file references in a content array.
Web search is being updated›
Web search is not recommended for near-term production workflows; avoid depending on it in production.
Thinking mode is always enabled›
K3 always has thinking enabled; control effort with the top-level reasoning_effort field using low, high or max.
No image input
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.